turboderp commited on
Commit
b0c6ec1
·
verified ·
1 Parent(s): cde055a

Upload folder using huggingface_hub

Browse files
.gitattributes CHANGED
@@ -33,3 +33,5 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
33
  *.zip filter=lfs diff=lfs merge=lfs -text
34
  *.zst filter=lfs diff=lfs merge=lfs -text
35
  *tfevents* filter=lfs diff=lfs merge=lfs -text
 
 
 
33
  *.zip filter=lfs diff=lfs merge=lfs -text
34
  *.zst filter=lfs diff=lfs merge=lfs -text
35
  *tfevents* filter=lfs diff=lfs merge=lfs -text
36
+ tekken.json filter=lfs diff=lfs merge=lfs -text
37
+ tokenizer.json filter=lfs diff=lfs merge=lfs -text
README.md CHANGED
@@ -1,285 +1,528 @@
1
  ---
 
 
 
 
 
 
 
 
 
 
 
 
 
2
  license: apache-2.0
3
- base_model: mistralai/Ministral-3-14B-Instruct-2512
4
- base_model_relation: quantized
5
- quantized_by: turboderp
 
 
 
6
  tags:
7
- - exl3
8
  ---
9
 
10
- EXL3 quants of [Ministral-3-14B-Instruct-2512](https://huggingface.co/mistralai/Ministral-3-14B-Instruct-2512)
11
-
12
- ⚠️ Requires ExLlamaV3 v0.0.17 (or v0.0.16 `dev` branch)
13
-
14
- [2.00 bits per weight](https://huggingface.co/turboderp/Ministral-3-14B-Instruct-2512-exl3/tree/2.00bpw)
15
- [2.25 bits per weight](https://huggingface.co/turboderp/Ministral-3-14B-Instruct-2512-exl3/tree/2.25bpw)
16
- [2.50 bits per weight](https://huggingface.co/turboderp/Ministral-3-14B-Instruct-2512-exl3/tree/2.50bpw)
17
- [3.00 bits per weight](https://huggingface.co/turboderp/Ministral-3-14B-Instruct-2512-exl3/tree/3.00bpw)
18
- [3.50 bits per weight](https://huggingface.co/turboderp/Ministral-3-14B-Instruct-2512-exl3/tree/3.50bpw)
19
- [4.00 bits per weight](https://huggingface.co/turboderp/Ministral-3-14B-Instruct-2512-exl3/tree/4.00bpw)
20
- [4.50 bits per weight](https://huggingface.co/turboderp/Ministral-3-14B-Instruct-2512-exl3/tree/4.50bpw)
21
- [5.00 bits per weight](https://huggingface.co/turboderp/Ministral-3-14B-Instruct-2512-exl3/tree/5.00bpw)
22
- [6.00 bits per weight](https://huggingface.co/turboderp/Ministral-3-14B-Instruct-2512-exl3/tree/6.00bpw)
23
- [7.00 bits per weight](https://huggingface.co/turboderp/Ministral-3-14B-Instruct-2512-exl3/tree/7.00bpw)
24
- [8.00 bits per weight](https://huggingface.co/turboderp/Ministral-3-14B-Instruct-2512-exl3/tree/8.00bpw)
25
-
26
- ![kld](https://cdn-uploads.huggingface.co/production/uploads/6383dc174c48969dcf1b4fce/jSa60ayyUZQlqER4-i5OL.png)
27
-
28
-
29
- # SVG Catbench
30
-
31
- <table>
32
- <tr>
33
- <td align="center">
34
- <a href="https://huggingface.co/turboderp/Ministral-3-14B-Instruct-2512-exl3/blob/main/2.00bpw.svg">
35
- <img src="2.00bpw.svg" alt="2.00 bpw" width="160">
36
- </a>
37
- <div>2.00 bpw</div>
38
- </td>
39
- <td align="center">
40
- <a href="https://huggingface.co/turboderp/Ministral-3-14B-Instruct-2512-exl3/blob/main/2.25bpw.svg">
41
- <img src="2.25bpw.svg" alt="2.25 bpw" width="160">
42
- </a>
43
- <div>2.25 bpw</div>
44
- </td>
45
- <td align="center">
46
- <a href="https://huggingface.co/turboderp/Ministral-3-14B-Instruct-2512-exl3/blob/main/2.50bpw.svg">
47
- <img src="2.50bpw.svg" alt="2.50 bpw" width="160">
48
- </a>
49
- <div>2.50 bpw</div>
50
- </td>
51
- <td align="center">
52
- <a href="https://huggingface.co/turboderp/Ministral-3-14B-Instruct-2512-exl3/blob/main/3.00bpw.svg">
53
- <img src="3.00bpw.svg" alt="3.00 bpw" width="160">
54
- </a>
55
- <div>3.00 bpw</div>
56
- </td>
57
- </tr>
58
- <tr>
59
- <td align="center">
60
- <a href="https://huggingface.co/tuboderp/Ministral-3-14B-Instruct-2512-exl3/blob/main/3.50bpw.svg">
61
- <img src="3.50bpw.svg" alt="3.50 bpw" width="160">
62
- </a>
63
- <div>3.50 bpw</div>
64
- </td>
65
- <td align="center">
66
- <a href="https://huggingface.co/turboderp/Ministral-3-14B-Instruct-2512-exl3/blob/main/4.00bpw.svg">
67
- <img src="4.00bpw.svg" alt="4.00 bpw" width="160">
68
- </a>
69
- <div>4.00 bpw</div>
70
- </td>
71
- <td align="center">
72
- <a href="https://huggingface.co/turboderp/Ministral-3-14B-Instruct-2512-exl3/blob/main/4.50bpw.svg">
73
- <img src="4.50bpw.svg" alt="4.50 bpw" width="160">
74
- </a>
75
- <div>4.50 bpw</div>
76
- </td>
77
- <td align="center">
78
- <a href="https://huggingface.co/turboderp/Ministral-3-14B-Instruct-2512-exl3/blob/main/5.00bpw.svg">
79
- <img src="5.00bpw.svg" alt="5.00 bpw" width="160">
80
- </a>
81
- <div>5.00 bpw</div>
82
- </td>
83
- </tr>
84
- <tr>
85
- <td align="center">
86
- <a href="https://huggingface.co/turboderp/Ministral-3-14B-Instruct-2512-exl3/blob/main/6.00bpw.svg">
87
- <img src="6.00bpw.svg" alt="6.00 bpw" width="160">
88
- </a>
89
- <div>6.00 bpw</div>
90
- </td>
91
- <td align="center">
92
- <a href="https://huggingface.co/turboderp/Ministral-3-14B-Instruct-2512-exl3/blob/main/7.00bpw.svg">
93
- <img src="7.00bpw.svg" alt="7.00 bpw" width="160">
94
- </a>
95
- <div>7.00 bpw</div>
96
- </td>
97
- <td align="center">
98
- <a href="https://huggingface.co/turboderp/Ministral-3-14B-Instruct-2512-exl3/blob/main/8.00bpw.svg">
99
- <img src="8.00bpw.svg" alt="8.00 bpw" width="160">
100
- </a>
101
- <div>8.00 bpw</div>
102
- </td>
103
- <td align="center">
104
- <a href="https://huggingface.co/turboderp/Ministral-3-14B-Instruct-2512-exl3/blob/main/hf.svg">
105
- <img src="hf.svg" alt="Transformers" width="160">
106
- </a>
107
- <div>Transformers</div>
108
- </td>
109
- </tr>
110
- </table>
111
- # Image captioning
112
-
113
- ![image](https://cdn-uploads.huggingface.co/production/uploads/6383dc174c48969dcf1b4fce/N1PxsxpUWDzt4B0Jd4Y1i.png)
114
 
115
- <details>
116
- <summary><b>2.00bpw</b></summary>
117
- This image depicts six cats perched atop and inside a vintage-style suitcase.
118
 
119
- The suitcase is grey with brown leather corners and accents, featuring a couple of golden hardware details and a classic handle. The cats appear to be of different varieties but all have distinct grey and black markings, giving them an adorable, harmonious look.
120
 
121
- - **Leftmost Cat:** This cat is grey and white, sitting atop the suitcase.
122
- - **Second Cat:** It has grey fur with a white face and chest, sitting at the edge of the suitcase.
123
- - **Third Cat:** This cat is grey, sitting inside the suitcase, near the center.
124
- - **Fourth Cat:** It has grey fur and a white chest, sitting closer to the center of the suitcase.
125
- - **Fifth Cat:** Another grey cat with a white chest, sitting right next to the fourth cat.
126
- - **Rightmost Cat:** This one is black and white, sitting outside the suitcase on the right.
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
127
 
128
- All the cats have varied expressions, some appearing curious or contemplative as they peer upward. The overall composition of the image gives a whimsical, playful vibe as the cats seem to be ready to embark on an adventure.
129
- </details>
130
- <details>
131
- <summary><b>2.25bpw</b></summary>
132
- The image shows six adorable kittens posed inside and around a vintage-style suitcase. Here’s a detailed description:
133
-
134
- 1. **Suitcase:**
135
- - The suitcase appears to be made of metal and has a classic, old-school design with a light brown leather strap and brass handles and fittings.
136
- - The interior is visible with the kittens sitting inside and one of them hanging its head out from the top opening.
137
-
138
- 2. **Kittens:**
139
- - The kittens have a variety of markings and colors.
140
- - Four of the kittens are gray with white markings on their faces and paws.
141
- - The fifth kitten is predominantly gray with a white face and a more distinct white chest area.
142
- - The last kitten on the right is mostly black with a white patch on its face.
143
- - The kittens are well-groomed, clean, and appear to be young and lively.
144
-
145
- 3. **Poses:**
146
- - Three kittens are positioned inside the suitcase, looking out and sitting on the surface.
147
- - The middle kitten is sticking its head out from the suitcase while its body is inside.
148
- - One kitten is sitting outside the suitcase, leaning against the edge.
149
- - The others are positioned both on top and beside the suitcase.
150
-
151
- The image is arranged in a playful and aesthetically pleasing manner, likely intended to evoke a sense of adventure and curiosity about where these kittens are going. It’s a charming composition, capturing the innocent and curious nature of kittens.
152
- </details>
153
- <details>
154
- <summary><b>2.50bpw</b></summary>
155
- This image showcases six cats situated in and around an open vintage-style suitcase. The scene appears to be staged for a creative and humorous purpose.
156
-
157
- From left to right:
158
- 1. A gray cat with a long tail and fluffy appearance is perched on the edge of the suitcase.
159
- 2. Another gray cat, also fluffy, is sitting just below the first cat, leaning out of the suitcase.
160
- 3. A gray cat with a white patch on its chest and a slightly raised posture is sitting inside the suitcase.
161
- 4. Another cat that looks similar to the third one but has a bit more white on its chest and face is also sitting inside the suitcase.
162
- 5. A gray cat with white markings on its face and chest, featuring a distinctive tufted tail, is positioned inside the suitcase with one paw resting on the edge and looking towards the camera.
163
- 6. A black cat with white paws is sitting inside the suitcase and gazing directly forward.
164
-
165
- The vintage suitcase has a sleek metallic frame and is adorned with brass studs and wooden accents. It is open, and the cats are positioned in a way that suggests they have just come out from inside it. The image creates a playful juxtaposition between the cats and the suitcase, perhaps evoking a sense of travel or adventure.
166
- </details>
167
  <details>
168
- <summary><b>3.00bpw</b></summary>
169
- This image shows six cats sitting together on and around an open suitcase.
170
-
171
- The cats are positioned in various ways:
172
- - Three cats are standing or sitting on the lid of the suitcase, which is a large, black rectangular container.
173
- - Two cats are positioned on the sides of the suitcase, one on the left and the other on the right.
174
- - The sixth cat is at the far right edge, sitting on the ground and partially leaning against the suitcase.
175
-
176
- The suitcase itself has a metallic, vintage look, with brass-colored handles and corners, and it has leather straps across the center of the lid. The background is plain and white, keeping the focus on the cats and the suitcase.
177
-
178
- The cats have different color patterns and coats:
179
- - One cat has a grey coat with a white chest and paws.
180
- - Another cat is grey with a white chest, but its face and paws are solid grey.
181
- - The third cat on the lid is entirely grey.
182
- - The fourth cat on the suitcase's edge has a grey face, white chest, and black paws and ears.
183
- - The fifth cat is predominantly grey with a black tail and patches on its body.
184
- - The last cat is entirely black.
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
185
  </details>
 
186
  <details>
187
- <summary><b>3.50bpw</b></summary>
188
- This image depicts a group of six cats posed in and around a large, vintage-style suitcase. The suitcase has a metallic exterior with brass-colored accents and brown leather corners.
189
-
190
- 1. **Cats:**
191
- - Five cats are perched on top of the suitcase, with one cat prominently sitting in the middle and looking directly at the camera.
192
- - This central cat has gray and white fur with yellowish eyes and is relatively large and fluffy.
193
- - To the left of the central cat, there are two gray cats, one with long fur and one with shorter fur, both also looking in the direction of the camera.
194
- - To the right of the central cat, there are two cats with similar gray and white fur, also looking towards the camera with bright yellow eyes.
195
- - One small black cat is standing to the far right, outside the suitcase and looking at the camera as well.
196
-
197
- 2. **Suitcase:**
198
- - The suitcase is open and appears quite spacious, accommodating multiple cats sitting inside.
199
- - It features several metal clasps and buckles, adding to its vintage aesthetic.
200
- - The exterior of the suitcase is predominantly silver with a dark brown leather trim and corners, giving it a distinctive, classic look.
201
-
202
- The background is plain white, which highlights the cats and the suitcase, making them the central focus of the image. The cats seem to be arranged in a playful manner, giving the image a whimsical and cute ambiance.
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
203
  </details>
 
204
  <details>
205
- <summary><b>4.00bpw</b></summary>
206
- The image shows six cats sitting and standing on and around a vintage-style suitcase. The cats seem to be of similar breeds and colors, likely siblings or closely related, as they share many physical traits. Here is a detailed breakdown of the cats and their positions:
207
 
208
- 1. The leftmost cat is sitting on top of the suitcase, with its front half outside the suitcase and its hind legs inside. It is grey and has a longhaired appearance.
209
- 2. The second cat from the left is perched on the edge of the suitcase with its front paws outside and its body inside the suitcase. It is grey with a white face, paws, and chest.
210
- 3. The third cat from the left is sitting comfortably inside the suitcase. This cat is similar in appearance to the leftmost cat with a grey longhaired coat.
211
- 4. The fourth cat from the left is also sitting inside the suitcase, just beside the third cat. It appears to have a greyish coat with a slightly darker hue compared to the others.
212
- 5. The fifth cat from the left is grey with a white face, paws, and chest, much like the second cat. It is sitting inside the suitcase with its body partially visible.
213
- 6. The rightmost cat is standing next to the suitcase with its full body outside. This cat is solid black in color, contrasting with the other grey cats.
214
 
215
- The suitcase itself is metallic with brown leather accents and several metal clasps and handles, giving it an old-fashioned, robust appearance. The setting appears to be clean and simple, possibly a studio, with the suitcase and cats being the main focus. The overall scene is quite whimsical and cute, resembling a playful or staged arrangement.
216
- </details>
217
- <details>
218
- <summary><b>4.50bpw</b></summary>
219
- This image features six cats sitting and standing atop a vintage-style suitcase. The suitcase appears to be an old-fashioned metal case with brass fittings and brown leather accents. Five of the cats have grey and white fur with long hair, and they are perched at various levels on the suitcase, creating a charming, whimsical composition. One grey-and-white cat is partially inside the open case, another sits on the lid, and three others are positioned on the edges and corners, looking alert and curious.
220
 
221
- The sixth cat is black with short hair and stands beside the suitcase on the ground, also looking towards the camera.
 
 
222
 
223
- The overall scene has a playful and somewhat staged look to it, highlighting the cats' natural curiosity and their penchant for climbing and sitting in boxes or confined spaces. The background is plain and white, which helps to keep the focus on the cats and their quirky setup.
224
- </details>
225
- <details>
226
- <summary><b>5.00bpw</b></summary>
227
- The image depicts six adorable kittens sitting and climbing on top of and inside a vintage-style suitcase. Here are the details:
228
 
229
- - The suitcase appears to be old and stylish, featuring brass buckles and corner guards, with leather straps and a tan handle.
230
- - There are five kittens on top of the suitcase. From left to right:
231
- 1. A light grey kitten with large golden eyes and a long, fluffy tail.
232
- 2. Another light grey kitten, similar in appearance to the first but with a slightly darker grey fur.
233
- 3. A grey tabby kitten with a distinctive white "M"-shaped marking on its forehead.
234
- 4. A grey kitten with a long, straight body and a short tail.
235
- 5. A darker grey kitten with bright yellow eyes, also with a distinctive white "M" shape on its forehead.
236
 
237
- - There is one black kitten sitting inside the suitcase, partially visible, with a white chest and paws, looking curious and alert.
 
238
 
239
- The kittens are all looking towards the camera, displaying various shades of grey fur and a mix of tabby patterns and solid colors. Their bright and expressive eyes contribute to the charming and endearing quality of the image.
240
- </details>
241
- <details>
242
- <summary><b>6.00bpw</b></summary>
243
- This image depicts six cats in a playful and arranged scene atop a vintage-style suitcase and a closed briefcase.
244
 
245
- 1. **Cats:**
246
- - There are four gray cats and two black-and-white cats.
247
- - The gray cats vary slightly in their shades, with some darker and some lighter.
248
- - The black-and-white cats have a mix of dark black and soft white fur.
249
- - They are all positioned in different poses, contributing to the dynamic composition.
 
250
 
251
- 2. **Suitcases:**
252
- - The cats are arranged on two pieces of luggage. One is a vintage-style open suitcase, and the other is a closed briefcase with a similar design.
253
- - The open suitcase has metal buckles and a handle, along with a fabric or leather interior.
254
- - The closed briefcase, which appears to be placed on top of the suitcase, also has metal buckles and a handle, with a dark exterior.
255
 
256
- 3. **Background:**
257
- - The background is plain and neutral, allowing the cats and the luggage to be the central focus of the image.
 
 
 
 
 
258
 
259
- 4. **Lighting and Composition:**
260
- - The lighting is even and bright, providing clear visibility of the cats and the luggage details.
261
- - The composition is balanced, with the cats positioned to create a visually appealing and engaging image, almost like a staged family portrait.
 
 
 
 
 
 
 
262
 
263
- This scene suggests a whimsical and light-hearted theme, often associated with cats and their penchant for finding cozy spots to relax. The vintage style of the suitcases adds a touch of charm to the image.
264
  </details>
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
265
  <details>
266
- <summary><b>7.00bpw</b></summary>
267
- The image shows a group of six kittens gathered around and climbing out of an open vintage-style suitcase. The suitcase appears to be made of metal with a sturdy design, featuring several brass-colored clasps, rivets, and a leather-like handle. Four of the kittens are inside the suitcase, with their heads and front paws visible, looking alert and curious. Two kittens are partially or fully outside of the suitcase, also showing a sense of curiosity and playfulness.
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
268
 
269
- The kittens are a mix of colors: three are gray, two have a mix of black and gray with white markings, and one is predominantly black with some white on its chest and legs. Their eyes are wide and golden, adding an endearing touch to the scene. The background is a plain, light-colored surface, which helps keep the focus on the kittens and the suitcase. The overall atmosphere of the image is charming and playful, capturing the curious nature of kittens.
270
  </details>
271
- <details>
272
- <summary><b>8.00bpw</b></summary>
273
- This image showcases a group of six kittens positioned atop and around two vintage-style suitcases.
274
 
275
- The suitcase on top, which is open, has a black exterior with metal corner handles and clasps, and is likely designed to evoke a sense of travel or adventure. The suitcase at the bottom is a bit smaller and has a similar vintage aesthetic with a combination of a light metal frame and wooden corners. There are brass clasps and decorative brass buttons on the leather straps.
276
 
277
- All the kittens seem to be in a playful and curious mood, exploring or climbing on the suitcases. They vary in color and pattern:
278
- - The first two kittens on the left are grey with longer fur.
279
- - The next two kittens are also grey but have shorter fur and a more solid appearance.
280
- - The kitten in the middle has a grey body with white paws and chest, and short fur.
281
- - The fifth kitten is grey with a black and white face.
282
- - The kitten on the far right is solid black.
283
 
284
- The background is plain and white, drawing attention to the kittens and the suitcases. The overall composition of the image is whimsical and charming, likely aiming to evoke a sense of warmth and joy.
285
- </details>
 
1
  ---
2
+ library_name: vllm
3
+ language:
4
+ - en
5
+ - fr
6
+ - es
7
+ - de
8
+ - it
9
+ - pt
10
+ - nl
11
+ - zh
12
+ - ja
13
+ - ko
14
+ - ar
15
  license: apache-2.0
16
+ inference: false
17
+ base_model:
18
+ - mistralai/Ministral-3-14B-Base-2512
19
+ extra_gated_description: >-
20
+ If you want to learn more about how we process your personal data, please read
21
+ our <a href="https://mistral.ai/terms/">Privacy Policy</a>.
22
  tags:
23
+ - mistral-common
24
  ---
25
 
26
+ # Ministral 3 14B Instruct 2512
27
+ The largest model in the Ministral 3 family, **Ministral 3 14B** offers frontier capabilities and performance comparable to its larger [Mistral Small 3.2 24B](https://huggingface.co/mistralai/Mistral-Small-3.2-Instruct-2506) counterpart. A powerful and efficient language model with vision capabilities.
28
+
29
+ This model is the instruct post-trained version in **FP8**, fine-tuned for instruction tasks, making it ideal for chat and instruction based use cases.
30
+
31
+ The Ministral 3 family is designed for edge deployment, capable of running on a wide range of hardware. Ministral 3 14B can even be deployed locally, capable of fitting in 24GB of VRAM in FP8, and less if further quantized.
32
+
33
+ Learn more in our blog post [here](https://mistral.ai/news/mistral-3).
34
+
35
+ ## Key Features
36
+ Ministral 3 14B consists of two main architectural components:
37
+ - **13.5B Language Model**
38
+ - **0.4B Vision Encoder**
39
+
40
+ The Ministral 3 14B Instruct model offers the following capabilities:
41
+ - **Vision**: Enables the model to analyze images and provide insights based on visual content, in addition to text.
42
+ - **Multilingual**: Supports dozens of languages, including English, French, Spanish, German, Italian, Portuguese, Dutch, Chinese, Japanese, Korean, Arabic.
43
+ - **System Prompt**: Maintains strong adherence and support for system prompts.
44
+ - **Agentic**: Offers best-in-class agentic capabilities with native function calling and JSON outputting.
45
+ - **Edge-Optimized**: Delivers best-in-class performance at a small scale, deployable anywhere.
46
+ - **Apache 2.0 License**: Open-source license allowing usage and modification for both commercial and non-commercial purposes.
47
+ - **Large Context Window**: Supports a 256k context window.
48
+
49
+ ### Use Cases
50
+ Private AI deployments where advanced capabilities meet practical hardware constraints:
51
+ - Private/custom chat and AI assistant deployments in constrained environments
52
+ - Advanced local agentic use cases
53
+ - Fine-tuning and specialization
54
+ - And more...
55
+
56
+ Bringing advanced AI capabilities to most environments.
57
+
58
+ ## Ministral 3 Family
59
+
60
+ | Model Name | Type | Precision | Link |
61
+ |--------------------------------|--------------------|-----------|------------------------------------------------------------------------------------------|
62
+ | Ministral 3 3B Base 2512 | Base pre-trained | BF16 | [Hugging Face](https://huggingface.co/mistralai/Ministral-3-3B-Base-2512) |
63
+ | Ministral 3 3B Instruct 2512 | Instruct post-trained | FP8 | [Hugging Face](https://huggingface.co/mistralai/Ministral-3-3B-Instruct-2512) |
64
+ | Ministral 3 3B Reasoning 2512 | Reasoning capable | BF16 | [Hugging Face](https://huggingface.co/mistralai/Ministral-3-3B-Reasoning-2512) |
65
+ | Ministral 3 8B Base 2512 | Base pre-trained | BF16 | [Hugging Face](https://huggingface.co/mistralai/Ministral-3-8B-Base-2512) |
66
+ | Ministral 3 8B Instruct 2512 | Instruct post-trained | FP8 | [Hugging Face](https://huggingface.co/mistralai/Ministral-3-8B-Instruct-2512) |
67
+ | Ministral 3 8B Reasoning 2512 | Reasoning capable | BF16 | [Hugging Face](https://huggingface.co/mistralai/Ministral-3-8B-Reasoning-2512) |
68
+ | Ministral 3 14B Base 2512 | Base pre-trained | BF16 | [Hugging Face](https://huggingface.co/mistralai/Ministral-3-14B-Base-2512) |
69
+ | **Ministral 3 14B Instruct 2512** | **Instruct post-trained** | **FP8** | [**Hugging Face**](https://huggingface.co/mistralai/Ministral-3-14B-Instruct-2512) |
70
+ | Ministral 3 14B Reasoning 2512 | Reasoning capable | BF16 | [Hugging Face](https://huggingface.co/mistralai/Ministral-3-14B-Reasoning-2512) |
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
71
 
72
+ Other formats available [here](https://huggingface.co/collections/mistralai/ministral-3-additional-checkpoints).
 
 
73
 
74
+ ## Benchmark Results
75
 
76
+ We compare Ministral 3 to similar sized models.
77
+
78
+ ### Reasoning
79
+
80
+ | Model | AIME25 | AIME24 | GPQA Diamond | LiveCodeBench |
81
+ |---------------------------|-------------|-------------|--------------|---------------|
82
+ | **Ministral 3 14B** | <u>0.850</u>| <u>0.898</u>| <u>0.712</u> | <u>0.646</u> |
83
+ | Qwen3-14B (Thinking) | 0.737 | 0.837 | 0.663 | 0.593 |
84
+ | | | | | |
85
+ | **Ministral 3 8B** | 0.787 | <u>0.860</u>| 0.668 | <u>0.616</u> |
86
+ | Qwen3-VL-8B-Thinking | <u>0.798</u>| <u>0.860</u>| <u>0.671</u> | 0.580 |
87
+ | | | | | |
88
+ | **Ministral 3 3B** | <u>0.721</u>| <u>0.775</u>| 0.534 | <u>0.548</u> |
89
+ | Qwen3-VL-4B-Thinking | 0.697 | 0.729 | <u>0.601</u> | 0.513 |
90
+
91
+ ### Instruct
92
+
93
+ | Model | Arena Hard | WildBench | MATH Maj@1 | MM MTBench |
94
+ |---------------------------|-------------|------------|-------------|------------------|
95
+ | **Ministral 3 14B** | <u>0.551</u>| <u>68.5</u>| <u>0.904</u>| <u>8.49</u> |
96
+ | Qwen3 14B (Non-Thinking) | 0.427 | 65.1 | 0.870 | NOT MULTIMODAL |
97
+ | Gemma3-12B-Instruct | 0.436 | 63.2 | 0.854 | 6.70 |
98
+ | | | | | |
99
+ | **Ministral 3 8B** | 0.509 | <u>66.8</u>| 0.876 | <u>8.08</u> |
100
+ | Qwen3-VL-8B-Instruct | <u>0.528</u>| 66.3 | <u>0.946</u>| 8.00 |
101
+ | | | | | |
102
+ | **Ministral 3 3B** | 0.305 | <u>56.8</u>| 0.830 | 7.83 |
103
+ | Qwen3-VL-4B-Instruct | <u>0.438</u>| <u>56.8</u>| <u>0.900</u>| <u>8.01</u> |
104
+ | Qwen3-VL-2B-Instruct | 0.163 | 42.2 | 0.786 | 6.36 |
105
+ | Gemma3-4B-Instruct | 0.318 | 49.1 | 0.759 | 5.23 |
106
+
107
+ ### Base
108
+
109
+ | Model | Multilingual MMLU | MATH CoT 2-Shot | AGIEval 5-shot | MMLU Redux 5-shot | MMLU 5-shot | TriviaQA 5-shot |
110
+ |---------------------|-------------------|-----------------|----------------|-------------------|-------------|-----------------|
111
+ | **Ministral 3 14B** | 0.742 | <u>0.676</u> | 0.648 | 0.820 | 0.794 | 0.749 |
112
+ | Qwen3 14B Base | <u>0.754</u> | 0.620 | <u>0.661</u> | <u>0.837</u> | <u>0.804</u>| 0.703 |
113
+ | Gemma 3 12B Base | 0.690 | 0.487 | 0.587 | 0.766 | 0.745 | <u>0.788</u> |
114
+ | | | | | | | |
115
+ | **Ministral 3 8B** | <u>0.706</u> | <u>0.626</u> | 0.591 | 0.793 | <u>0.761</u>| <u>0.681</u> |
116
+ | Qwen 3 8B Base | 0.700 | 0.576 | <u>0.596</u> | <u>0.794</u> | 0.760 | 0.639 |
117
+ | | | | | | | |
118
+ | **Ministral 3 3B** | 0.652 | <u>0.601</u> | 0.511 | 0.735 | 0.707 | 0.592 |
119
+ | Qwen 3 4B Base | <u>0.677</u> | 0.405 | <u>0.570</u> | <u>0.759</u> | <u>0.713</u>| 0.530 |
120
+ | Gemma 3 4B Base | 0.516 | 0.294 | 0.430 | 0.626 | 0.589 | <u>0.640</u> |
121
+
122
+ ## Usage
123
+
124
+ The model can be used with the following frameworks;
125
+ - [`vllm`](https://github.com/vllm-project/vllm): See [here](#vllm)
126
+ - [`transformers`](https://github.com/huggingface/transformers): See [here](#transformers)
127
+
128
+ ### vLLM
129
+
130
+ We recommend using this model with [vLLM](https://github.com/vllm-project/vllm).
131
+
132
+ #### Installation
133
+
134
+ Make sure to install most recent vllm:
135
+
136
+ ```
137
+ uv pip install -U vllm \
138
+ --torch-backend=auto \
139
+ --extra-index-url https://wheels.vllm.ai/nightly
140
+ ```
141
+
142
+ Doing so should automatically install [`mistral_common >= 1.8.6`](https://github.com/mistralai/mistral-common/releases/tag/v1.8.6).
143
+
144
+ To check:
145
+ ```
146
+ python -c "import mistral_common; print(mistral_common.__version__)"
147
+ ```
148
+
149
+ You can also make use of a ready-to-go [docker image](https://github.com/vllm-project/vllm/blob/main/Dockerfile) or on the [docker hub](https://hub.docker.com/layers/vllm/vllm-openai/latest/images/sha256-de9032a92ffea7b5c007dad80b38fd44aac11eddc31c435f8e52f3b7404bbf39).
150
+
151
+ #### Serve
152
+
153
+ Due to their size and the FP8 format of their weights `Ministral-3-3B-Instruct-2512`, `Ministral-3-8B-Instruct-2512` and `Ministral-3-14B-Instruct-2512` can run on a single 1xH200 GPU.
154
+
155
+ A simple launch command is:
156
+
157
+ ```bash
158
+ vllm serve mistralai/Ministral-3-14B-Instruct-2512 \
159
+ --tokenizer_mode mistral --config_format mistral --load_format mistral \
160
+ --enable-auto-tool-choice --tool-call-parser mistral
161
+ ```
162
+
163
+ Key parameter notes:
164
+
165
+ * enable-auto-tool-choice: Required when enabling tool usage.
166
+ * tool-call-parser mistral: Required when enabling tool usage.
167
+
168
+
169
+ Additional flags:
170
+
171
+ * You can set `--max-model-len` to preserve memory. By default it is set to `262144` which is quite large but not necessary for most scenarios.
172
+ * You can set `--max-num-batched-tokens` to balance throughput and latency, higher means higher throughput but higher latency.
173
+
174
+ #### Usage of the model
175
+
176
+ Here we asumme that the model `mistralai/Ministral-3-14B-Instruct-2512` is served and you can ping it to the domain `localhost` with the port `8000` which is the default for vLLM.
177
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
178
  <details>
179
+ <summary>Vision Reasoning</summary>
180
+
181
+ Let's see if the Ministral 3 knows when to pick a fight !
182
+
183
+ ```python
184
+ from datetime import datetime, timedelta
185
+
186
+ from openai import OpenAI
187
+ from huggingface_hub import hf_hub_download
188
+
189
+ # Modify OpenAI's API key and API base to use vLLM's API server.
190
+ openai_api_key = "EMPTY"
191
+ openai_api_base = "http://localhost:8000/v1"
192
+
193
+ TEMP = 0.15
194
+ MAX_TOK = 262144
195
+
196
+ client = OpenAI(
197
+ api_key=openai_api_key,
198
+ base_url=openai_api_base,
199
+ )
200
+
201
+ models = client.models.list()
202
+ model = models.data[0].id
203
+
204
+
205
+ def load_system_prompt(repo_id: str, filename: str) -> str:
206
+ file_path = hf_hub_download(repo_id=repo_id, filename=filename)
207
+ with open(file_path, "r") as file:
208
+ system_prompt = file.read()
209
+ today = datetime.today().strftime("%Y-%m-%d")
210
+ yesterday = (datetime.today() - timedelta(days=1)).strftime("%Y-%m-%d")
211
+ model_name = repo_id.split("/")[-1]
212
+ return system_prompt.format(name=model_name, today=today, yesterday=yesterday)
213
+
214
+
215
+ SYSTEM_PROMPT = load_system_prompt(model, "SYSTEM_PROMPT.txt")
216
+ image_url = "https://static.wikia.nocookie.net/essentialsdocs/images/7/70/Battle.png/revision/latest?cb=20220523172438"
217
+
218
+ messages = [
219
+ {"role": "system", "content": SYSTEM_PROMPT},
220
+ {
221
+ "role": "user",
222
+ "content": [
223
+ {
224
+ "type": "text",
225
+ "text": "What action do you think I should take in this situation? List all the possible actions and explain why you think they are good or bad.",
226
+ },
227
+ {"type": "image_url", "image_url": {"url": image_url}},
228
+ ],
229
+ },
230
+ ]
231
+
232
+
233
+ response = client.chat.completions.create(
234
+ model=model,
235
+ messages=messages,
236
+ temperature=TEMP,
237
+ max_tokens=MAX_TOK,
238
+ )
239
+
240
+ print(response.choices[0].message.content)
241
+ ```
242
+
243
  </details>
244
+
245
  <details>
246
+ <summary>Function Calling</summary>
247
+
248
+ Let's solve some equations thanks to our simple Python calculator tool.
249
+
250
+ ```python
251
+ import json
252
+ from openai import OpenAI
253
+ from huggingface_hub import hf_hub_download
254
+
255
+ # Modify OpenAI's API key and API base to use vLLM's API server.
256
+ openai_api_key = "EMPTY"
257
+ openai_api_base = "http://localhost:8000/v1"
258
+
259
+ TEMP = 0.15
260
+ MAX_TOK = 262144
261
+
262
+ client = OpenAI(
263
+ api_key=openai_api_key,
264
+ base_url=openai_api_base,
265
+ )
266
+
267
+ models = client.models.list()
268
+ model = models.data[0].id
269
+
270
+
271
+ def load_system_prompt(repo_id: str, filename: str) -> str:
272
+ file_path = hf_hub_download(repo_id=repo_id, filename=filename)
273
+ with open(file_path, "r") as file:
274
+ system_prompt = file.read()
275
+ return system_prompt
276
+
277
+
278
+ SYSTEM_PROMPT = load_system_prompt(model, "SYSTEM_PROMPT.txt")
279
+
280
+ image_url = "https://math-coaching.com/img/fiche/46/expressions-mathematiques.jpg"
281
+
282
+
283
+ def my_calculator(expression: str) -> str:
284
+ return str(eval(expression))
285
+
286
+
287
+ tools = [
288
+ {
289
+ "type": "function",
290
+ "function": {
291
+ "name": "my_calculator",
292
+ "description": "A calculator that can evaluate a mathematical expression.",
293
+ "parameters": {
294
+ "type": "object",
295
+ "properties": {
296
+ "expression": {
297
+ "type": "string",
298
+ "description": "The mathematical expression to evaluate.",
299
+ },
300
+ },
301
+ "required": ["expression"],
302
+ },
303
+ },
304
+ },
305
+ {
306
+ "type": "function",
307
+ "function": {
308
+ "name": "rewrite",
309
+ "description": "Rewrite a given text for improved clarity",
310
+ "parameters": {
311
+ "type": "object",
312
+ "properties": {
313
+ "text": {
314
+ "type": "string",
315
+ "description": "The input text to rewrite",
316
+ }
317
+ },
318
+ },
319
+ },
320
+ },
321
+ ]
322
+
323
+ messages = [
324
+ {"role": "system", "content": SYSTEM_PROMPT},
325
+ {
326
+ "role": "user",
327
+ "content": [
328
+ {
329
+ "type": "text",
330
+ "text": "Thanks to your calculator, compute the results for the equations that involve numbers displayed in the image.",
331
+ },
332
+ {
333
+ "type": "image_url",
334
+ "image_url": {
335
+ "url": image_url,
336
+ },
337
+ },
338
+ ],
339
+ },
340
+ ]
341
+
342
+ response = client.chat.completions.create(
343
+ model=model,
344
+ messages=messages,
345
+ temperature=TEMP,
346
+ max_tokens=MAX_TOK,
347
+ tools=tools,
348
+ tool_choice="auto",
349
+ )
350
+
351
+ tool_calls = response.choices[0].message.tool_calls
352
+
353
+ results = []
354
+ for tool_call in tool_calls:
355
+ function_name = tool_call.function.name
356
+ function_args = tool_call.function.arguments
357
+ if function_name == "my_calculator":
358
+ result = my_calculator(**json.loads(function_args))
359
+ results.append(result)
360
+
361
+ messages.append({"role": "assistant", "tool_calls": tool_calls})
362
+ for tool_call, result in zip(tool_calls, results):
363
+ messages.append(
364
+ {
365
+ "role": "tool",
366
+ "tool_call_id": tool_call.id,
367
+ "name": tool_call.function.name,
368
+ "content": result,
369
+ }
370
+ )
371
+
372
+
373
+ response = client.chat.completions.create(
374
+ model=model,
375
+ messages=messages,
376
+ temperature=TEMP,
377
+ max_tokens=MAX_TOK,
378
+ )
379
+
380
+ print(response.choices[0].message.content)
381
+ ```
382
+
383
  </details>
384
+
385
  <details>
386
+ <summary>Text-Only Request</summary>
 
387
 
388
+ Ministral 3 can follow your instructions to the letter.
 
 
 
 
 
389
 
390
+ ```python
391
+ from openai import OpenAI
392
+ from huggingface_hub import hf_hub_download
 
 
393
 
394
+ # Modify OpenAI's API key and API base to use vLLM's API server.
395
+ openai_api_key = "EMPTY"
396
+ openai_api_base = "http://localhost:8000/v1"
397
 
398
+ TEMP = 0.15
399
+ MAX_TOK = 262144
 
 
 
400
 
401
+ client = OpenAI(
402
+ api_key=openai_api_key,
403
+ base_url=openai_api_base,
404
+ )
 
 
 
405
 
406
+ models = client.models.list()
407
+ model = models.data[0].id
408
 
 
 
 
 
 
409
 
410
+ def load_system_prompt(repo_id: str, filename: str) -> str:
411
+ file_path = hf_hub_download(repo_id=repo_id, filename=filename)
412
+ with open(file_path, "r") as file:
413
+ system_prompt = file.read()
414
+ return system_prompt
415
+
416
 
417
+ SYSTEM_PROMPT = load_system_prompt(model, "SYSTEM_PROMPT.txt")
 
 
 
418
 
419
+ messages = [
420
+ {"role": "system", "content": SYSTEM_PROMPT},
421
+ {
422
+ "role": "user",
423
+ "content": "Write me a sentence where every word starts with the next letter in the alphabet - start with 'a' and end with 'z'.",
424
+ },
425
+ ]
426
 
427
+ response = client.chat.completions.create(
428
+ model=model,
429
+ messages=messages,
430
+ temperature=TEMP,
431
+ max_tokens=MAX_TOK,
432
+ )
433
+
434
+ assistant_message = response.choices[0].message.content
435
+ print(assistant_message)
436
+ ```
437
 
 
438
  </details>
439
+
440
+ ### Transformers
441
+
442
+ You can also use Ministral 3 14B Instruct 2512 with `Transformers` !
443
+
444
+ Transformers very recently added preliminary support for FP8, so please make sure to install from main:
445
+
446
+ ```sh
447
+ uv pip install git+https://github.com/huggingface/transformers
448
+ ```
449
+
450
+ To make the best use of our model with `Transformers` make sure to have [installed](https://github.com/mistralai/mistral-common) `mistral-common >= 1.8.6` to use our tokenizer.
451
+
452
+ ```bash
453
+ pip install mistral-common --upgrade
454
+ ```
455
+
456
+ Try it out by running the following snippet.
457
+
458
+ > [!Tip]
459
+ > By default Transformers will load the checkpoint in FP8 and dequantize it to BF16 on the fly,
460
+ > which means the model currently does not make use of accelerated FP8-kernels.
461
+ > Compatibility with accelerated FP8-kernels is currently worked on and will be available in a couple of weeks.
462
+ > Stay tuned!
463
+
464
  <details>
465
+ <summary>Python snippet</summary>
466
+
467
+ ```python
468
+ import torch
469
+ from transformers import Mistral3ForConditionalGeneration, MistralCommonBackend
470
+
471
+ model_id = "mistralai/Ministral-3-14B-Instruct-2512"
472
+
473
+ tokenizer = MistralCommonBackend.from_pretrained(model_id)
474
+ model = Mistral3ForConditionalGeneration.from_pretrained(model_id, device_map="auto")
475
+
476
+ image_url = "https://static.wikia.nocookie.net/essentialsdocs/images/7/70/Battle.png/revision/latest?cb=20220523172438"
477
+
478
+ messages = [
479
+ {
480
+ "role": "user",
481
+ "content": [
482
+ {
483
+ "type": "text",
484
+ "text": "What action do you think I should take in this situation? List all the possible actions and explain why you think they are good or bad.",
485
+ },
486
+ {"type": "image_url", "image_url": {"url": image_url}},
487
+ ],
488
+ },
489
+ ]
490
+
491
+ tokenized = tokenizer.apply_chat_template(messages, return_tensors="pt", return_dict=True)
492
+
493
+ tokenized["input_ids"] = tokenized["input_ids"].to(device="cuda")
494
+ tokenized["pixel_values"] = tokenized["pixel_values"].to(dtype=torch.bfloat16, device="cuda")
495
+ image_sizes = [tokenized["pixel_values"].shape[-2:]]
496
+
497
+ output = model.generate(
498
+ **tokenized,
499
+ image_sizes=image_sizes,
500
+ max_new_tokens=512,
501
+ )[0]
502
+
503
+ decoded_output = tokenizer.decode(output[len(tokenized["input_ids"][0]):])
504
+ print(decoded_output)
505
+ ```
506
+
507
+ **Note:**
508
+
509
+ Transformers allows you to automatically convert the checkpoint to Bfloat16. To so simple load the model as follows:
510
+
511
+ ```py
512
+ from transformers import Mistral3ForConditionalGeneration, FineGrainedFP8Config
513
+
514
+ model_id = "mistralai/Ministral-3-14B-Instruct-2512"
515
+ model = Mistral3ForConditionalGeneration.from_pretrained(
516
+ model_id,
517
+ device_map="auto",
518
+ quantization_config=FineGrainedFP8Config(dequantize=True)
519
+ )
520
+ ```
521
 
 
522
  </details>
 
 
 
523
 
524
+ ## License
525
 
526
+ This model is licensed under the [Apache 2.0 License](https://www.apache.org/licenses/LICENSE-2.0.txt).
 
 
 
 
 
527
 
528
+ *You must not use this model in a manner that infringes, misappropriates, or otherwise violates any third party’s rights, including intellectual property rights.*
 
SYSTEM_PROMPT.txt ADDED
@@ -0,0 +1,29 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ You are Ministral-3-14B-Instruct-2512, a Large Language Model (LLM) created by Mistral AI, a French startup headquartered in Paris.
2
+ You power an AI assistant called Le Chat.
3
+ Your knowledge base was last updated on 2023-10-01.
4
+ The current date is {today}.
5
+
6
+ When you're not sure about some information or when the user's request requires up-to-date or specific data, you must use the available tools to fetch the information. Do not hesitate to use tools whenever they can provide a more accurate or complete response. If no relevant tools are available, then clearly state that you don't have the information and avoid making up anything.
7
+ If the user's question is not clear, ambiguous, or does not provide enough context for you to accurately answer the question, you do not try to answer it right away and you rather ask the user to clarify their request (e.g. "What are some good restaurants around me?" => "Where are you?" or "When is the next flight to Tokyo" => "Where do you travel from?").
8
+ You are always very attentive to dates, in particular you try to resolve dates (e.g. "yesterday" is {yesterday}) and when asked about information at specific dates, you discard information that is at another date.
9
+ You follow these instructions in all languages, and always respond to the user in the language they use or request.
10
+ Next sections describe the capabilities that you have.
11
+
12
+ # WEB BROWSING INSTRUCTIONS
13
+
14
+ You cannot perform any web search or access internet to open URLs, links etc. If it seems like the user is expecting you to do so, you clarify the situation and ask the user to copy paste the text directly in the chat.
15
+
16
+ # MULTI-MODAL INSTRUCTIONS
17
+
18
+ You have the ability to read images, but you cannot generate images. You also cannot transcribe audio files or videos.
19
+ You cannot read nor transcribe audio files or videos.
20
+
21
+ # TOOL CALLING INSTRUCTIONS
22
+
23
+ You may have access to tools that you can use to fetch information or perform actions. You must use these tools in the following situations:
24
+
25
+ 1. When the request requires up-to-date information.
26
+ 2. When the request requires specific data that you do not have in your knowledge base.
27
+ 3. When the request involves actions that you cannot perform without tools.
28
+
29
+ Always prioritize using tools to provide the most accurate and helpful response. If tools are not available, inform the user that you cannot perform the requested action at the moment.
chat_template.jinja ADDED
@@ -0,0 +1,121 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {#- Default system message if no system prompt is passed. #}
2
+ {%- set default_system_message = 'You are Ministral-3-14B-Instruct-2512, a Large Language Model (LLM) created by Mistral AI, a French startup headquartered in Paris.\nYou power an AI assistant called Le Chat.\nYour knowledge base was last updated on 2023-10-01.\nThe current date is {today}.\n\nWhen you\'re not sure about some information or when the user\'s request requires up-to-date or specific data, you must use the available tools to fetch the information. Do not hesitate to use tools whenever they can provide a more accurate or complete response. If no relevant tools are available, then clearly state that you don\'t have the information and avoid making up anything.\nIf the user\'s question is not clear, ambiguous, or does not provide enough context for you to accurately answer the question, you do not try to answer it right away and you rather ask the user to clarify their request (e.g. "What are some good restaurants around me?" => "Where are you?" or "When is the next flight to Tokyo" => "Where do you travel from?").\nYou are always very attentive to dates, in particular you try to resolve dates (e.g. "yesterday" is {yesterday}) and when asked about information at specific dates, you discard information that is at another date.\nYou follow these instructions in all languages, and always respond to the user in the language they use or request.\nNext sections describe the capabilities that you have.\n\n# WEB BROWSING INSTRUCTIONS\n\nYou cannot perform any web search or access internet to open URLs, links etc. If it seems like the user is expecting you to do so, you clarify the situation and ask the user to copy paste the text directly in the chat.\n\n# MULTI-MODAL INSTRUCTIONS\n\nYou have the ability to read images, but you cannot generate images. You also cannot transcribe audio files or videos.\nYou cannot read nor transcribe audio files or videos.\n\n# TOOL CALLING INSTRUCTIONS\n\nYou may have access to tools that you can use to fetch information or perform actions. You must use these tools in the following situations:\n\n1. When the request requires up-to-date information.\n2. When the request requires specific data that you do not have in your knowledge base.\n3. When the request involves actions that you cannot perform without tools.\n\nAlways prioritize using tools to provide the most accurate and helpful response. If tools are not available, inform the user that you cannot perform the requested action at the moment.' %}
3
+
4
+ {#- Begin of sequence token. #}
5
+ {{- bos_token }}
6
+
7
+ {#- Handle system prompt if it exists. #}
8
+ {#- System prompt supports text content or text chunks. #}
9
+ {%- if messages[0]['role'] == 'system' %}
10
+ {{- '[SYSTEM_PROMPT]' -}}
11
+ {%- if messages[0]['content'] is string %}
12
+ {{- messages[0]['content'] -}}
13
+ {%- else %}
14
+ {%- for block in messages[0]['content'] %}
15
+ {%- if block['type'] == 'text' %}
16
+ {{- block['text'] }}
17
+ {%- else %}
18
+ {{- raise_exception('Only text chunks are supported in system message contents.') }}
19
+ {%- endif %}
20
+ {%- endfor %}
21
+ {%- endif %}
22
+ {{- '[/SYSTEM_PROMPT]' -}}
23
+ {%- set loop_messages = messages[1:] %}
24
+ {%- else %}
25
+ {%- set loop_messages = messages %}
26
+ {%- if default_system_message != '' %}
27
+ {{- '[SYSTEM_PROMPT]' + default_system_message + '[/SYSTEM_PROMPT]' }}
28
+ {%- endif %}
29
+ {%- endif %}
30
+
31
+
32
+ {#- Tools definition #}
33
+ {%- set tools_definition = '' %}
34
+ {%- set has_tools = false %}
35
+ {%- if tools is defined and tools is not none and tools|length > 0 %}
36
+ {%- set has_tools = true %}
37
+ {%- set tools_definition = '[AVAILABLE_TOOLS]' + (tools| tojson) + '[/AVAILABLE_TOOLS]' %}
38
+ {{- tools_definition }}
39
+ {%- endif %}
40
+
41
+ {#- Checks for alternating user/assistant messages. #}
42
+ {%- set ns = namespace(index=0) %}
43
+ {%- for message in loop_messages %}
44
+ {%- if message.role == 'user' or (message.role == 'assistant' and (message.tool_calls is not defined or message.tool_calls is none or message.tool_calls | length == 0)) %}
45
+ {%- if (message['role'] == 'user') != (ns.index % 2 == 0) %}
46
+ {{- raise_exception('After the optional system message, conversation roles must alternate user and assistant roles except for tool calls and results.') }}
47
+ {%- endif %}
48
+ {%- set ns.index = ns.index + 1 %}
49
+ {%- endif %}
50
+ {%- endfor %}
51
+
52
+ {#- Handle conversation messages. #}
53
+ {%- for message in loop_messages %}
54
+
55
+ {#- User messages supports text content or text and image chunks. #}
56
+ {%- if message['role'] == 'user' %}
57
+ {%- if message['content'] is string %}
58
+ {{- '[INST]' + message['content'] + '[/INST]' }}
59
+ {%- elif message['content'] | length > 0 %}
60
+ {{- '[INST]' }}
61
+ {%- if message['content'] | length == 2 %}
62
+ {%- set blocks = message['content'] | sort(attribute='type') %}
63
+ {%- else %}
64
+ {%- set blocks = message['content'] %}
65
+ {%- endif %}
66
+ {%- for block in blocks %}
67
+ {%- if block['type'] == 'text' %}
68
+ {{- block['text'] }}
69
+ {%- elif block['type'] in ['image', 'image_url'] %}
70
+ {{- '[IMG]' }}
71
+ {%- else %}
72
+ {{- raise_exception('Only text, image and image_url chunks are supported in user message content.') }}
73
+ {%- endif %}
74
+ {%- endfor %}
75
+ {{- '[/INST]' }}
76
+ {%- else %}
77
+ {{- raise_exception('User message must have a string or a list of chunks in content') }}
78
+ {%- endif %}
79
+
80
+ {#- Assistant messages supports text content or text and image chunks. #}
81
+ {%- elif message['role'] == 'assistant' %}
82
+ {%- if (message['content'] is none or message['content'] == '' or message['content']|length == 0) and (message['tool_calls'] is not defined or message['tool_calls'] is none or message['tool_calls']|length == 0) %}
83
+ {{- raise_exception('Assistant message must have a string or a list of chunks in content or a list of tool calls.') }}
84
+ {%- endif %}
85
+
86
+ {%- if message['content'] is string %}
87
+ {{- message['content'] }}
88
+ {%- elif message['content'] | length > 0 %}
89
+ {%- for block in message['content'] %}
90
+ {%- if block['type'] == 'text' %}
91
+ {{- block['text'] }}
92
+ {%- else %}
93
+ {{- raise_exception('Only text chunks are supported in assistant message contents.') }}
94
+ {%- endif %}
95
+ {%- endfor %}
96
+ {%- endif %}
97
+
98
+ {%- if message['tool_calls'] is defined and message['tool_calls'] is not none and message['tool_calls']|length > 0 %}
99
+ {%- for tool in message['tool_calls'] %}
100
+ {%- set arguments = tool['function']['arguments'] %}
101
+ {%- if arguments is not string %}
102
+ {%- set arguments = arguments|tojson|safe %}
103
+ {%- elif arguments == '' %}
104
+ {%- set arguments = '{}' %}
105
+ {%- endif %}
106
+ {{- '[TOOL_CALLS]' + tool['function']['name'] + '[ARGS]' + arguments }}
107
+ {%- endfor %}
108
+ {%- endif %}
109
+
110
+ {#- End of sequence token for each assistant messages. #}
111
+ {{- eos_token }}
112
+
113
+ {#- Tool messages only supports text content. #}
114
+ {%- elif message['role'] == 'tool' %}
115
+ {{- '[TOOL_RESULTS]' + message['content']|string + '[/TOOL_RESULTS]' }}
116
+
117
+ {#- Raise exception for unsupported roles. #}
118
+ {%- else %}
119
+ {{- raise_exception('Only user, assistant and tool roles are supported, got ' + message['role'] + '.') }}
120
+ {%- endif %}
121
+ {%- endfor %}
config.json ADDED
@@ -0,0 +1,87 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "architectures": [
3
+ "Mistral3ForConditionalGeneration"
4
+ ],
5
+ "dtype": "bfloat16",
6
+ "image_token_index": 10,
7
+ "tie_word_embeddings": false,
8
+ "model_type": "mistral3",
9
+ "multimodal_projector_bias": false,
10
+ "projector_hidden_act": "gelu",
11
+ "quantization_config": {
12
+ "quant_method": "exl3",
13
+ "version": "0.0.16",
14
+ "bits": 8.0,
15
+ "head_bits": 6,
16
+ "calibration": {
17
+ "rows": 250,
18
+ "cols": 2048
19
+ },
20
+ "out_scales": "always",
21
+ "codebook": "mcg",
22
+ "original_quantization_config": {
23
+ "activation_scheme": "static",
24
+ "dequantize": false,
25
+ "modules_to_not_convert": [
26
+ "model.vision_tower",
27
+ "model.multi_modal_projector",
28
+ "lm_head",
29
+ "model.vision_tower",
30
+ "model.multi_modal_projector",
31
+ "lm_head"
32
+ ],
33
+ "quant_method": "fp8",
34
+ "weight_block_size": null
35
+ }
36
+ },
37
+ "spatial_merge_size": 2,
38
+ "text_config": {
39
+ "attention_dropout": 0.0,
40
+ "head_dim": 128,
41
+ "hidden_act": "silu",
42
+ "hidden_size": 5120,
43
+ "initializer_range": 0.02,
44
+ "intermediate_size": 16384,
45
+ "max_position_embeddings": 262144,
46
+ "model_type": "ministral3",
47
+ "num_attention_heads": 32,
48
+ "num_hidden_layers": 40,
49
+ "num_key_value_heads": 8,
50
+ "rms_norm_eps": 1e-05,
51
+ "rope_parameters": {
52
+ "beta_fast": 32.0,
53
+ "beta_slow": 1.0,
54
+ "factor": 16.0,
55
+ "llama_4_scaling_beta": 0.1,
56
+ "mscale": 1.0,
57
+ "mscale_all_dim": 1.0,
58
+ "original_max_position_embeddings": 16384,
59
+ "rope_theta": 1000000000.0,
60
+ "rope_type": "yarn",
61
+ "type": "yarn"
62
+ },
63
+ "sliding_window": null,
64
+ "use_cache": true,
65
+ "vocab_size": 131072
66
+ },
67
+ "transformers_version": "5.0.0.dev0",
68
+ "vision_config": {
69
+ "attention_dropout": 0.0,
70
+ "head_dim": 64,
71
+ "hidden_act": "silu",
72
+ "hidden_size": 1024,
73
+ "image_size": 1540,
74
+ "initializer_range": 0.02,
75
+ "intermediate_size": 4096,
76
+ "model_type": "pixtral",
77
+ "num_attention_heads": 16,
78
+ "num_channels": 3,
79
+ "num_hidden_layers": 24,
80
+ "patch_size": 14,
81
+ "rope_parameters": {
82
+ "rope_theta": 10000.0,
83
+ "rope_type": "default"
84
+ }
85
+ },
86
+ "vision_feature_layer": -1
87
+ }
generation_config.json ADDED
@@ -0,0 +1,7 @@
 
 
 
 
 
 
 
 
1
+ {
2
+ "bos_token_id": 1,
3
+ "eos_token_id": 2,
4
+ "max_length": 262144,
5
+ "pad_token_id": 11,
6
+ "transformers_version": "5.0.0.dev0"
7
+ }
model-00001-of-00002.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:b2d5010137ee1bfeee53225eabf48c92a13b31421564b30e6c45c2c31a7aa202
3
+ size 8341115140
model-00002-of-00002.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:c550b94ced4bf0ec76521645a5f6d87af697dfded5e0a867a1eea3fc66b1847f
3
+ size 6554672936
model.safetensors.index.json ADDED
The diff for this file is too large to render. See raw diff
 
params.json ADDED
@@ -0,0 +1,51 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "dim": 5120,
3
+ "n_layers": 40,
4
+ "head_dim": 128,
5
+ "hidden_dim": 16384,
6
+ "n_heads": 32,
7
+ "n_kv_heads": 8,
8
+ "rope_theta": 1000000000.0,
9
+ "norm_eps": 1e-05,
10
+ "vocab_size": 131072,
11
+ "tied_embeddings": false,
12
+ "max_position_embeddings": 262144,
13
+ "llama_4_scaling": {
14
+ "original_max_position_embeddings": 16384,
15
+ "beta": 0.1
16
+ },
17
+ "q_lora_rank": null,
18
+ "qk_rope_head_dim": null,
19
+ "qk_nope_head_dim": null,
20
+ "kv_lora_rank": null,
21
+ "v_head_dim": null,
22
+ "quantization": {
23
+ "qformat_weight": "fp8_e4m3",
24
+ "qscheme_act": "TENSOR"
25
+ },
26
+ "yarn": {
27
+ "original_max_position_embeddings": 16384,
28
+ "factor": 16,
29
+ "apply_scale": false,
30
+ "beta": 32,
31
+ "alpha": 1
32
+ },
33
+ "vision_encoder": {
34
+ "image_token_id": 10,
35
+ "image_break_token_id": 12,
36
+ "image_end_token_id": 13,
37
+ "intermediate_size": 4096,
38
+ "num_hidden_layers": 24,
39
+ "num_attention_heads": 16,
40
+ "mm_projector_id": "patch_merge",
41
+ "spatial_merge_size": 2,
42
+ "hidden_size": 1024,
43
+ "num_channels": 3,
44
+ "image_size": 1540,
45
+ "max_image_size": 1540,
46
+ "patch_size": 14,
47
+ "rope_theta": 10000.0,
48
+ "add_pre_mm_projector_layer_norm": true,
49
+ "adapter_bias": false
50
+ }
51
+ }
preprocessor_config.json ADDED
@@ -0,0 +1,34 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "crop_size": null,
3
+ "data_format": "channels_first",
4
+ "device": null,
5
+ "disable_grouping": null,
6
+ "do_center_crop": null,
7
+ "do_convert_rgb": true,
8
+ "do_normalize": true,
9
+ "do_pad": null,
10
+ "do_rescale": true,
11
+ "do_resize": true,
12
+ "image_mean": [
13
+ 0.48145466,
14
+ 0.4578275,
15
+ 0.40821073
16
+ ],
17
+ "image_processor_type": "PixtralImageProcessorFast",
18
+ "image_seq_length": null,
19
+ "image_std": [
20
+ 0.26862954,
21
+ 0.26130258,
22
+ 0.27577711
23
+ ],
24
+ "input_data_format": null,
25
+ "pad_size": null,
26
+ "patch_size": 14,
27
+ "processor_class": "PixtralProcessor",
28
+ "resample": 3,
29
+ "rescale_factor": 0.00392156862745098,
30
+ "return_tensors": null,
31
+ "size": {
32
+ "longest_edge": 1540
33
+ }
34
+ }
processor_config.json ADDED
@@ -0,0 +1,42 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "image_break_token": "[IMG_BREAK]",
3
+ "image_end_token": "[IMG_END]",
4
+ "image_processor": {
5
+ "crop_size": null,
6
+ "data_format": "channels_first",
7
+ "device": null,
8
+ "disable_grouping": null,
9
+ "do_center_crop": null,
10
+ "do_convert_rgb": true,
11
+ "do_normalize": true,
12
+ "do_pad": null,
13
+ "do_rescale": true,
14
+ "do_resize": true,
15
+ "image_mean": [
16
+ 0.48145466,
17
+ 0.4578275,
18
+ 0.40821073
19
+ ],
20
+ "image_processor_type": "PixtralImageProcessorFast",
21
+ "image_seq_length": null,
22
+ "image_std": [
23
+ 0.26862954,
24
+ 0.26130258,
25
+ 0.27577711
26
+ ],
27
+ "input_data_format": null,
28
+ "pad_size": null,
29
+ "patch_size": 14,
30
+ "processor_class": "PixtralProcessor",
31
+ "resample": 3,
32
+ "rescale_factor": 0.00392156862745098,
33
+ "return_tensors": null,
34
+ "size": {
35
+ "longest_edge": 1540
36
+ }
37
+ },
38
+ "image_token": "[IMG]",
39
+ "patch_size": 14,
40
+ "processor_class": "PixtralProcessor",
41
+ "spatial_merge_size": 2
42
+ }
quantization_config.json ADDED
The diff for this file is too large to render. See raw diff
 
tekken.json ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:e29d19ea32eb7e26e6c0572d57cb7f9eca0f4420e0e0fe6ae1cf3be94da1c0d6
3
+ size 16753777
tokenizer.json ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:286acad9b0e27fce778ac429763536accf618ccb6ed72963b6f94685e531c5c7
3
+ size 17077402
tokenizer_config.json ADDED
@@ -0,0 +1,1010 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "additional_special_tokens": null,
3
+ "backend": "tokenizers",
4
+ "extra_special_tokens": [
5
+ "<unk>",
6
+ "<s>",
7
+ "</s>",
8
+ "[INST]",
9
+ "[/INST]",
10
+ "[AVAILABLE_TOOLS]",
11
+ "[/AVAILABLE_TOOLS]",
12
+ "[TOOL_RESULTS]",
13
+ "[/TOOL_RESULTS]",
14
+ "[TOOL_CALLS]",
15
+ "[IMG]",
16
+ "<pad>",
17
+ "[IMG_BREAK]",
18
+ "[IMG_END]",
19
+ "[PREFIX]",
20
+ "[MIDDLE]",
21
+ "[SUFFIX]",
22
+ "[SYSTEM_PROMPT]",
23
+ "[/SYSTEM_PROMPT]",
24
+ "[TOOL_CONTENT]",
25
+ "<SPECIAL_20>",
26
+ "<SPECIAL_21>",
27
+ "<SPECIAL_22>",
28
+ "<SPECIAL_23>",
29
+ "[AUDIO]",
30
+ "[BEGIN_AUDIO]",
31
+ "<SPECIAL_26>",
32
+ "<SPECIAL_27>",
33
+ "<SPECIAL_28>",
34
+ "<SPECIAL_29>",
35
+ "<SPECIAL_30>",
36
+ "<SPECIAL_31>",
37
+ "[ARGS]",
38
+ "[CALL_ID]",
39
+ "[THINK]",
40
+ "[/THINK]",
41
+ "<SPECIAL_36>",
42
+ "<SPECIAL_37>",
43
+ "<SPECIAL_38>",
44
+ "<SPECIAL_39>",
45
+ "<SPECIAL_40>",
46
+ "<SPECIAL_41>",
47
+ "<SPECIAL_42>",
48
+ "<SPECIAL_43>",
49
+ "<SPECIAL_44>",
50
+ "<SPECIAL_45>",
51
+ "<SPECIAL_46>",
52
+ "<SPECIAL_47>",
53
+ "<SPECIAL_48>",
54
+ "<SPECIAL_49>",
55
+ "<SPECIAL_50>",
56
+ "<SPECIAL_51>",
57
+ "<SPECIAL_52>",
58
+ "<SPECIAL_53>",
59
+ "<SPECIAL_54>",
60
+ "<SPECIAL_55>",
61
+ "<SPECIAL_56>",
62
+ "<SPECIAL_57>",
63
+ "<SPECIAL_58>",
64
+ "<SPECIAL_59>",
65
+ "<SPECIAL_60>",
66
+ "<SPECIAL_61>",
67
+ "<SPECIAL_62>",
68
+ "<SPECIAL_63>",
69
+ "<SPECIAL_64>",
70
+ "<SPECIAL_65>",
71
+ "<SPECIAL_66>",
72
+ "<SPECIAL_67>",
73
+ "<SPECIAL_68>",
74
+ "<SPECIAL_69>",
75
+ "<SPECIAL_70>",
76
+ "<SPECIAL_71>",
77
+ "<SPECIAL_72>",
78
+ "<SPECIAL_73>",
79
+ "<SPECIAL_74>",
80
+ "<SPECIAL_75>",
81
+ "<SPECIAL_76>",
82
+ "<SPECIAL_77>",
83
+ "<SPECIAL_78>",
84
+ "<SPECIAL_79>",
85
+ "<SPECIAL_80>",
86
+ "<SPECIAL_81>",
87
+ "<SPECIAL_82>",
88
+ "<SPECIAL_83>",
89
+ "<SPECIAL_84>",
90
+ "<SPECIAL_85>",
91
+ "<SPECIAL_86>",
92
+ "<SPECIAL_87>",
93
+ "<SPECIAL_88>",
94
+ "<SPECIAL_89>",
95
+ "<SPECIAL_90>",
96
+ "<SPECIAL_91>",
97
+ "<SPECIAL_92>",
98
+ "<SPECIAL_93>",
99
+ "<SPECIAL_94>",
100
+ "<SPECIAL_95>",
101
+ "<SPECIAL_96>",
102
+ "<SPECIAL_97>",
103
+ "<SPECIAL_98>",
104
+ "<SPECIAL_99>",
105
+ "<SPECIAL_100>",
106
+ "<SPECIAL_101>",
107
+ "<SPECIAL_102>",
108
+ "<SPECIAL_103>",
109
+ "<SPECIAL_104>",
110
+ "<SPECIAL_105>",
111
+ "<SPECIAL_106>",
112
+ "<SPECIAL_107>",
113
+ "<SPECIAL_108>",
114
+ "<SPECIAL_109>",
115
+ "<SPECIAL_110>",
116
+ "<SPECIAL_111>",
117
+ "<SPECIAL_112>",
118
+ "<SPECIAL_113>",
119
+ "<SPECIAL_114>",
120
+ "<SPECIAL_115>",
121
+ "<SPECIAL_116>",
122
+ "<SPECIAL_117>",
123
+ "<SPECIAL_118>",
124
+ "<SPECIAL_119>",
125
+ "<SPECIAL_120>",
126
+ "<SPECIAL_121>",
127
+ "<SPECIAL_122>",
128
+ "<SPECIAL_123>",
129
+ "<SPECIAL_124>",
130
+ "<SPECIAL_125>",
131
+ "<SPECIAL_126>",
132
+ "<SPECIAL_127>",
133
+ "<SPECIAL_128>",
134
+ "<SPECIAL_129>",
135
+ "<SPECIAL_130>",
136
+ "<SPECIAL_131>",
137
+ "<SPECIAL_132>",
138
+ "<SPECIAL_133>",
139
+ "<SPECIAL_134>",
140
+ "<SPECIAL_135>",
141
+ "<SPECIAL_136>",
142
+ "<SPECIAL_137>",
143
+ "<SPECIAL_138>",
144
+ "<SPECIAL_139>",
145
+ "<SPECIAL_140>",
146
+ "<SPECIAL_141>",
147
+ "<SPECIAL_142>",
148
+ "<SPECIAL_143>",
149
+ "<SPECIAL_144>",
150
+ "<SPECIAL_145>",
151
+ "<SPECIAL_146>",
152
+ "<SPECIAL_147>",
153
+ "<SPECIAL_148>",
154
+ "<SPECIAL_149>",
155
+ "<SPECIAL_150>",
156
+ "<SPECIAL_151>",
157
+ "<SPECIAL_152>",
158
+ "<SPECIAL_153>",
159
+ "<SPECIAL_154>",
160
+ "<SPECIAL_155>",
161
+ "<SPECIAL_156>",
162
+ "<SPECIAL_157>",
163
+ "<SPECIAL_158>",
164
+ "<SPECIAL_159>",
165
+ "<SPECIAL_160>",
166
+ "<SPECIAL_161>",
167
+ "<SPECIAL_162>",
168
+ "<SPECIAL_163>",
169
+ "<SPECIAL_164>",
170
+ "<SPECIAL_165>",
171
+ "<SPECIAL_166>",
172
+ "<SPECIAL_167>",
173
+ "<SPECIAL_168>",
174
+ "<SPECIAL_169>",
175
+ "<SPECIAL_170>",
176
+ "<SPECIAL_171>",
177
+ "<SPECIAL_172>",
178
+ "<SPECIAL_173>",
179
+ "<SPECIAL_174>",
180
+ "<SPECIAL_175>",
181
+ "<SPECIAL_176>",
182
+ "<SPECIAL_177>",
183
+ "<SPECIAL_178>",
184
+ "<SPECIAL_179>",
185
+ "<SPECIAL_180>",
186
+ "<SPECIAL_181>",
187
+ "<SPECIAL_182>",
188
+ "<SPECIAL_183>",
189
+ "<SPECIAL_184>",
190
+ "<SPECIAL_185>",
191
+ "<SPECIAL_186>",
192
+ "<SPECIAL_187>",
193
+ "<SPECIAL_188>",
194
+ "<SPECIAL_189>",
195
+ "<SPECIAL_190>",
196
+ "<SPECIAL_191>",
197
+ "<SPECIAL_192>",
198
+ "<SPECIAL_193>",
199
+ "<SPECIAL_194>",
200
+ "<SPECIAL_195>",
201
+ "<SPECIAL_196>",
202
+ "<SPECIAL_197>",
203
+ "<SPECIAL_198>",
204
+ "<SPECIAL_199>",
205
+ "<SPECIAL_200>",
206
+ "<SPECIAL_201>",
207
+ "<SPECIAL_202>",
208
+ "<SPECIAL_203>",
209
+ "<SPECIAL_204>",
210
+ "<SPECIAL_205>",
211
+ "<SPECIAL_206>",
212
+ "<SPECIAL_207>",
213
+ "<SPECIAL_208>",
214
+ "<SPECIAL_209>",
215
+ "<SPECIAL_210>",
216
+ "<SPECIAL_211>",
217
+ "<SPECIAL_212>",
218
+ "<SPECIAL_213>",
219
+ "<SPECIAL_214>",
220
+ "<SPECIAL_215>",
221
+ "<SPECIAL_216>",
222
+ "<SPECIAL_217>",
223
+ "<SPECIAL_218>",
224
+ "<SPECIAL_219>",
225
+ "<SPECIAL_220>",
226
+ "<SPECIAL_221>",
227
+ "<SPECIAL_222>",
228
+ "<SPECIAL_223>",
229
+ "<SPECIAL_224>",
230
+ "<SPECIAL_225>",
231
+ "<SPECIAL_226>",
232
+ "<SPECIAL_227>",
233
+ "<SPECIAL_228>",
234
+ "<SPECIAL_229>",
235
+ "<SPECIAL_230>",
236
+ "<SPECIAL_231>",
237
+ "<SPECIAL_232>",
238
+ "<SPECIAL_233>",
239
+ "<SPECIAL_234>",
240
+ "<SPECIAL_235>",
241
+ "<SPECIAL_236>",
242
+ "<SPECIAL_237>",
243
+ "<SPECIAL_238>",
244
+ "<SPECIAL_239>",
245
+ "<SPECIAL_240>",
246
+ "<SPECIAL_241>",
247
+ "<SPECIAL_242>",
248
+ "<SPECIAL_243>",
249
+ "<SPECIAL_244>",
250
+ "<SPECIAL_245>",
251
+ "<SPECIAL_246>",
252
+ "<SPECIAL_247>",
253
+ "<SPECIAL_248>",
254
+ "<SPECIAL_249>",
255
+ "<SPECIAL_250>",
256
+ "<SPECIAL_251>",
257
+ "<SPECIAL_252>",
258
+ "<SPECIAL_253>",
259
+ "<SPECIAL_254>",
260
+ "<SPECIAL_255>",
261
+ "<SPECIAL_256>",
262
+ "<SPECIAL_257>",
263
+ "<SPECIAL_258>",
264
+ "<SPECIAL_259>",
265
+ "<SPECIAL_260>",
266
+ "<SPECIAL_261>",
267
+ "<SPECIAL_262>",
268
+ "<SPECIAL_263>",
269
+ "<SPECIAL_264>",
270
+ "<SPECIAL_265>",
271
+ "<SPECIAL_266>",
272
+ "<SPECIAL_267>",
273
+ "<SPECIAL_268>",
274
+ "<SPECIAL_269>",
275
+ "<SPECIAL_270>",
276
+ "<SPECIAL_271>",
277
+ "<SPECIAL_272>",
278
+ "<SPECIAL_273>",
279
+ "<SPECIAL_274>",
280
+ "<SPECIAL_275>",
281
+ "<SPECIAL_276>",
282
+ "<SPECIAL_277>",
283
+ "<SPECIAL_278>",
284
+ "<SPECIAL_279>",
285
+ "<SPECIAL_280>",
286
+ "<SPECIAL_281>",
287
+ "<SPECIAL_282>",
288
+ "<SPECIAL_283>",
289
+ "<SPECIAL_284>",
290
+ "<SPECIAL_285>",
291
+ "<SPECIAL_286>",
292
+ "<SPECIAL_287>",
293
+ "<SPECIAL_288>",
294
+ "<SPECIAL_289>",
295
+ "<SPECIAL_290>",
296
+ "<SPECIAL_291>",
297
+ "<SPECIAL_292>",
298
+ "<SPECIAL_293>",
299
+ "<SPECIAL_294>",
300
+ "<SPECIAL_295>",
301
+ "<SPECIAL_296>",
302
+ "<SPECIAL_297>",
303
+ "<SPECIAL_298>",
304
+ "<SPECIAL_299>",
305
+ "<SPECIAL_300>",
306
+ "<SPECIAL_301>",
307
+ "<SPECIAL_302>",
308
+ "<SPECIAL_303>",
309
+ "<SPECIAL_304>",
310
+ "<SPECIAL_305>",
311
+ "<SPECIAL_306>",
312
+ "<SPECIAL_307>",
313
+ "<SPECIAL_308>",
314
+ "<SPECIAL_309>",
315
+ "<SPECIAL_310>",
316
+ "<SPECIAL_311>",
317
+ "<SPECIAL_312>",
318
+ "<SPECIAL_313>",
319
+ "<SPECIAL_314>",
320
+ "<SPECIAL_315>",
321
+ "<SPECIAL_316>",
322
+ "<SPECIAL_317>",
323
+ "<SPECIAL_318>",
324
+ "<SPECIAL_319>",
325
+ "<SPECIAL_320>",
326
+ "<SPECIAL_321>",
327
+ "<SPECIAL_322>",
328
+ "<SPECIAL_323>",
329
+ "<SPECIAL_324>",
330
+ "<SPECIAL_325>",
331
+ "<SPECIAL_326>",
332
+ "<SPECIAL_327>",
333
+ "<SPECIAL_328>",
334
+ "<SPECIAL_329>",
335
+ "<SPECIAL_330>",
336
+ "<SPECIAL_331>",
337
+ "<SPECIAL_332>",
338
+ "<SPECIAL_333>",
339
+ "<SPECIAL_334>",
340
+ "<SPECIAL_335>",
341
+ "<SPECIAL_336>",
342
+ "<SPECIAL_337>",
343
+ "<SPECIAL_338>",
344
+ "<SPECIAL_339>",
345
+ "<SPECIAL_340>",
346
+ "<SPECIAL_341>",
347
+ "<SPECIAL_342>",
348
+ "<SPECIAL_343>",
349
+ "<SPECIAL_344>",
350
+ "<SPECIAL_345>",
351
+ "<SPECIAL_346>",
352
+ "<SPECIAL_347>",
353
+ "<SPECIAL_348>",
354
+ "<SPECIAL_349>",
355
+ "<SPECIAL_350>",
356
+ "<SPECIAL_351>",
357
+ "<SPECIAL_352>",
358
+ "<SPECIAL_353>",
359
+ "<SPECIAL_354>",
360
+ "<SPECIAL_355>",
361
+ "<SPECIAL_356>",
362
+ "<SPECIAL_357>",
363
+ "<SPECIAL_358>",
364
+ "<SPECIAL_359>",
365
+ "<SPECIAL_360>",
366
+ "<SPECIAL_361>",
367
+ "<SPECIAL_362>",
368
+ "<SPECIAL_363>",
369
+ "<SPECIAL_364>",
370
+ "<SPECIAL_365>",
371
+ "<SPECIAL_366>",
372
+ "<SPECIAL_367>",
373
+ "<SPECIAL_368>",
374
+ "<SPECIAL_369>",
375
+ "<SPECIAL_370>",
376
+ "<SPECIAL_371>",
377
+ "<SPECIAL_372>",
378
+ "<SPECIAL_373>",
379
+ "<SPECIAL_374>",
380
+ "<SPECIAL_375>",
381
+ "<SPECIAL_376>",
382
+ "<SPECIAL_377>",
383
+ "<SPECIAL_378>",
384
+ "<SPECIAL_379>",
385
+ "<SPECIAL_380>",
386
+ "<SPECIAL_381>",
387
+ "<SPECIAL_382>",
388
+ "<SPECIAL_383>",
389
+ "<SPECIAL_384>",
390
+ "<SPECIAL_385>",
391
+ "<SPECIAL_386>",
392
+ "<SPECIAL_387>",
393
+ "<SPECIAL_388>",
394
+ "<SPECIAL_389>",
395
+ "<SPECIAL_390>",
396
+ "<SPECIAL_391>",
397
+ "<SPECIAL_392>",
398
+ "<SPECIAL_393>",
399
+ "<SPECIAL_394>",
400
+ "<SPECIAL_395>",
401
+ "<SPECIAL_396>",
402
+ "<SPECIAL_397>",
403
+ "<SPECIAL_398>",
404
+ "<SPECIAL_399>",
405
+ "<SPECIAL_400>",
406
+ "<SPECIAL_401>",
407
+ "<SPECIAL_402>",
408
+ "<SPECIAL_403>",
409
+ "<SPECIAL_404>",
410
+ "<SPECIAL_405>",
411
+ "<SPECIAL_406>",
412
+ "<SPECIAL_407>",
413
+ "<SPECIAL_408>",
414
+ "<SPECIAL_409>",
415
+ "<SPECIAL_410>",
416
+ "<SPECIAL_411>",
417
+ "<SPECIAL_412>",
418
+ "<SPECIAL_413>",
419
+ "<SPECIAL_414>",
420
+ "<SPECIAL_415>",
421
+ "<SPECIAL_416>",
422
+ "<SPECIAL_417>",
423
+ "<SPECIAL_418>",
424
+ "<SPECIAL_419>",
425
+ "<SPECIAL_420>",
426
+ "<SPECIAL_421>",
427
+ "<SPECIAL_422>",
428
+ "<SPECIAL_423>",
429
+ "<SPECIAL_424>",
430
+ "<SPECIAL_425>",
431
+ "<SPECIAL_426>",
432
+ "<SPECIAL_427>",
433
+ "<SPECIAL_428>",
434
+ "<SPECIAL_429>",
435
+ "<SPECIAL_430>",
436
+ "<SPECIAL_431>",
437
+ "<SPECIAL_432>",
438
+ "<SPECIAL_433>",
439
+ "<SPECIAL_434>",
440
+ "<SPECIAL_435>",
441
+ "<SPECIAL_436>",
442
+ "<SPECIAL_437>",
443
+ "<SPECIAL_438>",
444
+ "<SPECIAL_439>",
445
+ "<SPECIAL_440>",
446
+ "<SPECIAL_441>",
447
+ "<SPECIAL_442>",
448
+ "<SPECIAL_443>",
449
+ "<SPECIAL_444>",
450
+ "<SPECIAL_445>",
451
+ "<SPECIAL_446>",
452
+ "<SPECIAL_447>",
453
+ "<SPECIAL_448>",
454
+ "<SPECIAL_449>",
455
+ "<SPECIAL_450>",
456
+ "<SPECIAL_451>",
457
+ "<SPECIAL_452>",
458
+ "<SPECIAL_453>",
459
+ "<SPECIAL_454>",
460
+ "<SPECIAL_455>",
461
+ "<SPECIAL_456>",
462
+ "<SPECIAL_457>",
463
+ "<SPECIAL_458>",
464
+ "<SPECIAL_459>",
465
+ "<SPECIAL_460>",
466
+ "<SPECIAL_461>",
467
+ "<SPECIAL_462>",
468
+ "<SPECIAL_463>",
469
+ "<SPECIAL_464>",
470
+ "<SPECIAL_465>",
471
+ "<SPECIAL_466>",
472
+ "<SPECIAL_467>",
473
+ "<SPECIAL_468>",
474
+ "<SPECIAL_469>",
475
+ "<SPECIAL_470>",
476
+ "<SPECIAL_471>",
477
+ "<SPECIAL_472>",
478
+ "<SPECIAL_473>",
479
+ "<SPECIAL_474>",
480
+ "<SPECIAL_475>",
481
+ "<SPECIAL_476>",
482
+ "<SPECIAL_477>",
483
+ "<SPECIAL_478>",
484
+ "<SPECIAL_479>",
485
+ "<SPECIAL_480>",
486
+ "<SPECIAL_481>",
487
+ "<SPECIAL_482>",
488
+ "<SPECIAL_483>",
489
+ "<SPECIAL_484>",
490
+ "<SPECIAL_485>",
491
+ "<SPECIAL_486>",
492
+ "<SPECIAL_487>",
493
+ "<SPECIAL_488>",
494
+ "<SPECIAL_489>",
495
+ "<SPECIAL_490>",
496
+ "<SPECIAL_491>",
497
+ "<SPECIAL_492>",
498
+ "<SPECIAL_493>",
499
+ "<SPECIAL_494>",
500
+ "<SPECIAL_495>",
501
+ "<SPECIAL_496>",
502
+ "<SPECIAL_497>",
503
+ "<SPECIAL_498>",
504
+ "<SPECIAL_499>",
505
+ "<SPECIAL_500>",
506
+ "<SPECIAL_501>",
507
+ "<SPECIAL_502>",
508
+ "<SPECIAL_503>",
509
+ "<SPECIAL_504>",
510
+ "<SPECIAL_505>",
511
+ "<SPECIAL_506>",
512
+ "<SPECIAL_507>",
513
+ "<SPECIAL_508>",
514
+ "<SPECIAL_509>",
515
+ "<SPECIAL_510>",
516
+ "<SPECIAL_511>",
517
+ "<SPECIAL_512>",
518
+ "<SPECIAL_513>",
519
+ "<SPECIAL_514>",
520
+ "<SPECIAL_515>",
521
+ "<SPECIAL_516>",
522
+ "<SPECIAL_517>",
523
+ "<SPECIAL_518>",
524
+ "<SPECIAL_519>",
525
+ "<SPECIAL_520>",
526
+ "<SPECIAL_521>",
527
+ "<SPECIAL_522>",
528
+ "<SPECIAL_523>",
529
+ "<SPECIAL_524>",
530
+ "<SPECIAL_525>",
531
+ "<SPECIAL_526>",
532
+ "<SPECIAL_527>",
533
+ "<SPECIAL_528>",
534
+ "<SPECIAL_529>",
535
+ "<SPECIAL_530>",
536
+ "<SPECIAL_531>",
537
+ "<SPECIAL_532>",
538
+ "<SPECIAL_533>",
539
+ "<SPECIAL_534>",
540
+ "<SPECIAL_535>",
541
+ "<SPECIAL_536>",
542
+ "<SPECIAL_537>",
543
+ "<SPECIAL_538>",
544
+ "<SPECIAL_539>",
545
+ "<SPECIAL_540>",
546
+ "<SPECIAL_541>",
547
+ "<SPECIAL_542>",
548
+ "<SPECIAL_543>",
549
+ "<SPECIAL_544>",
550
+ "<SPECIAL_545>",
551
+ "<SPECIAL_546>",
552
+ "<SPECIAL_547>",
553
+ "<SPECIAL_548>",
554
+ "<SPECIAL_549>",
555
+ "<SPECIAL_550>",
556
+ "<SPECIAL_551>",
557
+ "<SPECIAL_552>",
558
+ "<SPECIAL_553>",
559
+ "<SPECIAL_554>",
560
+ "<SPECIAL_555>",
561
+ "<SPECIAL_556>",
562
+ "<SPECIAL_557>",
563
+ "<SPECIAL_558>",
564
+ "<SPECIAL_559>",
565
+ "<SPECIAL_560>",
566
+ "<SPECIAL_561>",
567
+ "<SPECIAL_562>",
568
+ "<SPECIAL_563>",
569
+ "<SPECIAL_564>",
570
+ "<SPECIAL_565>",
571
+ "<SPECIAL_566>",
572
+ "<SPECIAL_567>",
573
+ "<SPECIAL_568>",
574
+ "<SPECIAL_569>",
575
+ "<SPECIAL_570>",
576
+ "<SPECIAL_571>",
577
+ "<SPECIAL_572>",
578
+ "<SPECIAL_573>",
579
+ "<SPECIAL_574>",
580
+ "<SPECIAL_575>",
581
+ "<SPECIAL_576>",
582
+ "<SPECIAL_577>",
583
+ "<SPECIAL_578>",
584
+ "<SPECIAL_579>",
585
+ "<SPECIAL_580>",
586
+ "<SPECIAL_581>",
587
+ "<SPECIAL_582>",
588
+ "<SPECIAL_583>",
589
+ "<SPECIAL_584>",
590
+ "<SPECIAL_585>",
591
+ "<SPECIAL_586>",
592
+ "<SPECIAL_587>",
593
+ "<SPECIAL_588>",
594
+ "<SPECIAL_589>",
595
+ "<SPECIAL_590>",
596
+ "<SPECIAL_591>",
597
+ "<SPECIAL_592>",
598
+ "<SPECIAL_593>",
599
+ "<SPECIAL_594>",
600
+ "<SPECIAL_595>",
601
+ "<SPECIAL_596>",
602
+ "<SPECIAL_597>",
603
+ "<SPECIAL_598>",
604
+ "<SPECIAL_599>",
605
+ "<SPECIAL_600>",
606
+ "<SPECIAL_601>",
607
+ "<SPECIAL_602>",
608
+ "<SPECIAL_603>",
609
+ "<SPECIAL_604>",
610
+ "<SPECIAL_605>",
611
+ "<SPECIAL_606>",
612
+ "<SPECIAL_607>",
613
+ "<SPECIAL_608>",
614
+ "<SPECIAL_609>",
615
+ "<SPECIAL_610>",
616
+ "<SPECIAL_611>",
617
+ "<SPECIAL_612>",
618
+ "<SPECIAL_613>",
619
+ "<SPECIAL_614>",
620
+ "<SPECIAL_615>",
621
+ "<SPECIAL_616>",
622
+ "<SPECIAL_617>",
623
+ "<SPECIAL_618>",
624
+ "<SPECIAL_619>",
625
+ "<SPECIAL_620>",
626
+ "<SPECIAL_621>",
627
+ "<SPECIAL_622>",
628
+ "<SPECIAL_623>",
629
+ "<SPECIAL_624>",
630
+ "<SPECIAL_625>",
631
+ "<SPECIAL_626>",
632
+ "<SPECIAL_627>",
633
+ "<SPECIAL_628>",
634
+ "<SPECIAL_629>",
635
+ "<SPECIAL_630>",
636
+ "<SPECIAL_631>",
637
+ "<SPECIAL_632>",
638
+ "<SPECIAL_633>",
639
+ "<SPECIAL_634>",
640
+ "<SPECIAL_635>",
641
+ "<SPECIAL_636>",
642
+ "<SPECIAL_637>",
643
+ "<SPECIAL_638>",
644
+ "<SPECIAL_639>",
645
+ "<SPECIAL_640>",
646
+ "<SPECIAL_641>",
647
+ "<SPECIAL_642>",
648
+ "<SPECIAL_643>",
649
+ "<SPECIAL_644>",
650
+ "<SPECIAL_645>",
651
+ "<SPECIAL_646>",
652
+ "<SPECIAL_647>",
653
+ "<SPECIAL_648>",
654
+ "<SPECIAL_649>",
655
+ "<SPECIAL_650>",
656
+ "<SPECIAL_651>",
657
+ "<SPECIAL_652>",
658
+ "<SPECIAL_653>",
659
+ "<SPECIAL_654>",
660
+ "<SPECIAL_655>",
661
+ "<SPECIAL_656>",
662
+ "<SPECIAL_657>",
663
+ "<SPECIAL_658>",
664
+ "<SPECIAL_659>",
665
+ "<SPECIAL_660>",
666
+ "<SPECIAL_661>",
667
+ "<SPECIAL_662>",
668
+ "<SPECIAL_663>",
669
+ "<SPECIAL_664>",
670
+ "<SPECIAL_665>",
671
+ "<SPECIAL_666>",
672
+ "<SPECIAL_667>",
673
+ "<SPECIAL_668>",
674
+ "<SPECIAL_669>",
675
+ "<SPECIAL_670>",
676
+ "<SPECIAL_671>",
677
+ "<SPECIAL_672>",
678
+ "<SPECIAL_673>",
679
+ "<SPECIAL_674>",
680
+ "<SPECIAL_675>",
681
+ "<SPECIAL_676>",
682
+ "<SPECIAL_677>",
683
+ "<SPECIAL_678>",
684
+ "<SPECIAL_679>",
685
+ "<SPECIAL_680>",
686
+ "<SPECIAL_681>",
687
+ "<SPECIAL_682>",
688
+ "<SPECIAL_683>",
689
+ "<SPECIAL_684>",
690
+ "<SPECIAL_685>",
691
+ "<SPECIAL_686>",
692
+ "<SPECIAL_687>",
693
+ "<SPECIAL_688>",
694
+ "<SPECIAL_689>",
695
+ "<SPECIAL_690>",
696
+ "<SPECIAL_691>",
697
+ "<SPECIAL_692>",
698
+ "<SPECIAL_693>",
699
+ "<SPECIAL_694>",
700
+ "<SPECIAL_695>",
701
+ "<SPECIAL_696>",
702
+ "<SPECIAL_697>",
703
+ "<SPECIAL_698>",
704
+ "<SPECIAL_699>",
705
+ "<SPECIAL_700>",
706
+ "<SPECIAL_701>",
707
+ "<SPECIAL_702>",
708
+ "<SPECIAL_703>",
709
+ "<SPECIAL_704>",
710
+ "<SPECIAL_705>",
711
+ "<SPECIAL_706>",
712
+ "<SPECIAL_707>",
713
+ "<SPECIAL_708>",
714
+ "<SPECIAL_709>",
715
+ "<SPECIAL_710>",
716
+ "<SPECIAL_711>",
717
+ "<SPECIAL_712>",
718
+ "<SPECIAL_713>",
719
+ "<SPECIAL_714>",
720
+ "<SPECIAL_715>",
721
+ "<SPECIAL_716>",
722
+ "<SPECIAL_717>",
723
+ "<SPECIAL_718>",
724
+ "<SPECIAL_719>",
725
+ "<SPECIAL_720>",
726
+ "<SPECIAL_721>",
727
+ "<SPECIAL_722>",
728
+ "<SPECIAL_723>",
729
+ "<SPECIAL_724>",
730
+ "<SPECIAL_725>",
731
+ "<SPECIAL_726>",
732
+ "<SPECIAL_727>",
733
+ "<SPECIAL_728>",
734
+ "<SPECIAL_729>",
735
+ "<SPECIAL_730>",
736
+ "<SPECIAL_731>",
737
+ "<SPECIAL_732>",
738
+ "<SPECIAL_733>",
739
+ "<SPECIAL_734>",
740
+ "<SPECIAL_735>",
741
+ "<SPECIAL_736>",
742
+ "<SPECIAL_737>",
743
+ "<SPECIAL_738>",
744
+ "<SPECIAL_739>",
745
+ "<SPECIAL_740>",
746
+ "<SPECIAL_741>",
747
+ "<SPECIAL_742>",
748
+ "<SPECIAL_743>",
749
+ "<SPECIAL_744>",
750
+ "<SPECIAL_745>",
751
+ "<SPECIAL_746>",
752
+ "<SPECIAL_747>",
753
+ "<SPECIAL_748>",
754
+ "<SPECIAL_749>",
755
+ "<SPECIAL_750>",
756
+ "<SPECIAL_751>",
757
+ "<SPECIAL_752>",
758
+ "<SPECIAL_753>",
759
+ "<SPECIAL_754>",
760
+ "<SPECIAL_755>",
761
+ "<SPECIAL_756>",
762
+ "<SPECIAL_757>",
763
+ "<SPECIAL_758>",
764
+ "<SPECIAL_759>",
765
+ "<SPECIAL_760>",
766
+ "<SPECIAL_761>",
767
+ "<SPECIAL_762>",
768
+ "<SPECIAL_763>",
769
+ "<SPECIAL_764>",
770
+ "<SPECIAL_765>",
771
+ "<SPECIAL_766>",
772
+ "<SPECIAL_767>",
773
+ "<SPECIAL_768>",
774
+ "<SPECIAL_769>",
775
+ "<SPECIAL_770>",
776
+ "<SPECIAL_771>",
777
+ "<SPECIAL_772>",
778
+ "<SPECIAL_773>",
779
+ "<SPECIAL_774>",
780
+ "<SPECIAL_775>",
781
+ "<SPECIAL_776>",
782
+ "<SPECIAL_777>",
783
+ "<SPECIAL_778>",
784
+ "<SPECIAL_779>",
785
+ "<SPECIAL_780>",
786
+ "<SPECIAL_781>",
787
+ "<SPECIAL_782>",
788
+ "<SPECIAL_783>",
789
+ "<SPECIAL_784>",
790
+ "<SPECIAL_785>",
791
+ "<SPECIAL_786>",
792
+ "<SPECIAL_787>",
793
+ "<SPECIAL_788>",
794
+ "<SPECIAL_789>",
795
+ "<SPECIAL_790>",
796
+ "<SPECIAL_791>",
797
+ "<SPECIAL_792>",
798
+ "<SPECIAL_793>",
799
+ "<SPECIAL_794>",
800
+ "<SPECIAL_795>",
801
+ "<SPECIAL_796>",
802
+ "<SPECIAL_797>",
803
+ "<SPECIAL_798>",
804
+ "<SPECIAL_799>",
805
+ "<SPECIAL_800>",
806
+ "<SPECIAL_801>",
807
+ "<SPECIAL_802>",
808
+ "<SPECIAL_803>",
809
+ "<SPECIAL_804>",
810
+ "<SPECIAL_805>",
811
+ "<SPECIAL_806>",
812
+ "<SPECIAL_807>",
813
+ "<SPECIAL_808>",
814
+ "<SPECIAL_809>",
815
+ "<SPECIAL_810>",
816
+ "<SPECIAL_811>",
817
+ "<SPECIAL_812>",
818
+ "<SPECIAL_813>",
819
+ "<SPECIAL_814>",
820
+ "<SPECIAL_815>",
821
+ "<SPECIAL_816>",
822
+ "<SPECIAL_817>",
823
+ "<SPECIAL_818>",
824
+ "<SPECIAL_819>",
825
+ "<SPECIAL_820>",
826
+ "<SPECIAL_821>",
827
+ "<SPECIAL_822>",
828
+ "<SPECIAL_823>",
829
+ "<SPECIAL_824>",
830
+ "<SPECIAL_825>",
831
+ "<SPECIAL_826>",
832
+ "<SPECIAL_827>",
833
+ "<SPECIAL_828>",
834
+ "<SPECIAL_829>",
835
+ "<SPECIAL_830>",
836
+ "<SPECIAL_831>",
837
+ "<SPECIAL_832>",
838
+ "<SPECIAL_833>",
839
+ "<SPECIAL_834>",
840
+ "<SPECIAL_835>",
841
+ "<SPECIAL_836>",
842
+ "<SPECIAL_837>",
843
+ "<SPECIAL_838>",
844
+ "<SPECIAL_839>",
845
+ "<SPECIAL_840>",
846
+ "<SPECIAL_841>",
847
+ "<SPECIAL_842>",
848
+ "<SPECIAL_843>",
849
+ "<SPECIAL_844>",
850
+ "<SPECIAL_845>",
851
+ "<SPECIAL_846>",
852
+ "<SPECIAL_847>",
853
+ "<SPECIAL_848>",
854
+ "<SPECIAL_849>",
855
+ "<SPECIAL_850>",
856
+ "<SPECIAL_851>",
857
+ "<SPECIAL_852>",
858
+ "<SPECIAL_853>",
859
+ "<SPECIAL_854>",
860
+ "<SPECIAL_855>",
861
+ "<SPECIAL_856>",
862
+ "<SPECIAL_857>",
863
+ "<SPECIAL_858>",
864
+ "<SPECIAL_859>",
865
+ "<SPECIAL_860>",
866
+ "<SPECIAL_861>",
867
+ "<SPECIAL_862>",
868
+ "<SPECIAL_863>",
869
+ "<SPECIAL_864>",
870
+ "<SPECIAL_865>",
871
+ "<SPECIAL_866>",
872
+ "<SPECIAL_867>",
873
+ "<SPECIAL_868>",
874
+ "<SPECIAL_869>",
875
+ "<SPECIAL_870>",
876
+ "<SPECIAL_871>",
877
+ "<SPECIAL_872>",
878
+ "<SPECIAL_873>",
879
+ "<SPECIAL_874>",
880
+ "<SPECIAL_875>",
881
+ "<SPECIAL_876>",
882
+ "<SPECIAL_877>",
883
+ "<SPECIAL_878>",
884
+ "<SPECIAL_879>",
885
+ "<SPECIAL_880>",
886
+ "<SPECIAL_881>",
887
+ "<SPECIAL_882>",
888
+ "<SPECIAL_883>",
889
+ "<SPECIAL_884>",
890
+ "<SPECIAL_885>",
891
+ "<SPECIAL_886>",
892
+ "<SPECIAL_887>",
893
+ "<SPECIAL_888>",
894
+ "<SPECIAL_889>",
895
+ "<SPECIAL_890>",
896
+ "<SPECIAL_891>",
897
+ "<SPECIAL_892>",
898
+ "<SPECIAL_893>",
899
+ "<SPECIAL_894>",
900
+ "<SPECIAL_895>",
901
+ "<SPECIAL_896>",
902
+ "<SPECIAL_897>",
903
+ "<SPECIAL_898>",
904
+ "<SPECIAL_899>",
905
+ "<SPECIAL_900>",
906
+ "<SPECIAL_901>",
907
+ "<SPECIAL_902>",
908
+ "<SPECIAL_903>",
909
+ "<SPECIAL_904>",
910
+ "<SPECIAL_905>",
911
+ "<SPECIAL_906>",
912
+ "<SPECIAL_907>",
913
+ "<SPECIAL_908>",
914
+ "<SPECIAL_909>",
915
+ "<SPECIAL_910>",
916
+ "<SPECIAL_911>",
917
+ "<SPECIAL_912>",
918
+ "<SPECIAL_913>",
919
+ "<SPECIAL_914>",
920
+ "<SPECIAL_915>",
921
+ "<SPECIAL_916>",
922
+ "<SPECIAL_917>",
923
+ "<SPECIAL_918>",
924
+ "<SPECIAL_919>",
925
+ "<SPECIAL_920>",
926
+ "<SPECIAL_921>",
927
+ "<SPECIAL_922>",
928
+ "<SPECIAL_923>",
929
+ "<SPECIAL_924>",
930
+ "<SPECIAL_925>",
931
+ "<SPECIAL_926>",
932
+ "<SPECIAL_927>",
933
+ "<SPECIAL_928>",
934
+ "<SPECIAL_929>",
935
+ "<SPECIAL_930>",
936
+ "<SPECIAL_931>",
937
+ "<SPECIAL_932>",
938
+ "<SPECIAL_933>",
939
+ "<SPECIAL_934>",
940
+ "<SPECIAL_935>",
941
+ "<SPECIAL_936>",
942
+ "<SPECIAL_937>",
943
+ "<SPECIAL_938>",
944
+ "<SPECIAL_939>",
945
+ "<SPECIAL_940>",
946
+ "<SPECIAL_941>",
947
+ "<SPECIAL_942>",
948
+ "<SPECIAL_943>",
949
+ "<SPECIAL_944>",
950
+ "<SPECIAL_945>",
951
+ "<SPECIAL_946>",
952
+ "<SPECIAL_947>",
953
+ "<SPECIAL_948>",
954
+ "<SPECIAL_949>",
955
+ "<SPECIAL_950>",
956
+ "<SPECIAL_951>",
957
+ "<SPECIAL_952>",
958
+ "<SPECIAL_953>",
959
+ "<SPECIAL_954>",
960
+ "<SPECIAL_955>",
961
+ "<SPECIAL_956>",
962
+ "<SPECIAL_957>",
963
+ "<SPECIAL_958>",
964
+ "<SPECIAL_959>",
965
+ "<SPECIAL_960>",
966
+ "<SPECIAL_961>",
967
+ "<SPECIAL_962>",
968
+ "<SPECIAL_963>",
969
+ "<SPECIAL_964>",
970
+ "<SPECIAL_965>",
971
+ "<SPECIAL_966>",
972
+ "<SPECIAL_967>",
973
+ "<SPECIAL_968>",
974
+ "<SPECIAL_969>",
975
+ "<SPECIAL_970>",
976
+ "<SPECIAL_971>",
977
+ "<SPECIAL_972>",
978
+ "<SPECIAL_973>",
979
+ "<SPECIAL_974>",
980
+ "<SPECIAL_975>",
981
+ "<SPECIAL_976>",
982
+ "<SPECIAL_977>",
983
+ "<SPECIAL_978>",
984
+ "<SPECIAL_979>",
985
+ "<SPECIAL_980>",
986
+ "<SPECIAL_981>",
987
+ "<SPECIAL_982>",
988
+ "<SPECIAL_983>",
989
+ "<SPECIAL_984>",
990
+ "<SPECIAL_985>",
991
+ "<SPECIAL_986>",
992
+ "<SPECIAL_987>",
993
+ "<SPECIAL_988>",
994
+ "<SPECIAL_989>",
995
+ "<SPECIAL_990>",
996
+ "<SPECIAL_991>",
997
+ "<SPECIAL_992>",
998
+ "<SPECIAL_993>",
999
+ "<SPECIAL_994>",
1000
+ "<SPECIAL_995>",
1001
+ "<SPECIAL_996>",
1002
+ "<SPECIAL_997>",
1003
+ "<SPECIAL_998>",
1004
+ "<SPECIAL_999>"
1005
+ ],
1006
+ "model_max_length": 1000000000000000019884624838656,
1007
+ "pad_token": "<pad>",
1008
+ "processor_class": "PixtralProcessor",
1009
+ "tokenizer_class": "TokenizersBackend"
1010
+ }